Innovation, Dierentiation, and the Choice of an Underwriter: Evidence from Equity Linked Securities

Enrique Schroth HEC - University of Lausanney and FAME First draft: February 10th, 2003. First Revision: April 17th, 2004.

am grateful to the members of my thesis committee, Douglas Gale, Boyan Jovanovic, and Marti Subrahmanyam. I also acknowledge helpful comments by Franklin Allen, Chris Flinn, Helios Herrera, Eugenio Miravete, Franco Peracchi, Michael Rockinger and Peter Tufano. Any errors in the paper are mine. I also thank Ingo Walter, Iftekhar Hasan and the Berkley Center for Entrepreneurial Studies who made possible my access to the data. y Address: Route de Chavannes 33, 1007 Lausanne, Switzerland. E-mail: enrique.schroth@hec.unil.ch. Tel.: +41 21 692 3346. Fax +41 21 692 3435.

Innovation, Dierentiation, and the Choice of an Underwriter: Evidence from Equity Linked Securities

Abstract Investment banks imitate other banks innovative corporate securities with their own varieties, and compete with the innovator to underwrite new issues. This paper uses data of all the corporate oerings of Equity-Linked and Derivative Securities from the SDC records to estimate the issuers demand of underwriting services provided by investment banks across dierent varieties of securities. The results show that, ceteris paribus, the demand for the innovators variety is larger than for the imitators. The estimated demand advantage is decreasing in time, and it decreases faster if the security appeared later in a sequence of innovations. JEL classication: G24, C25, L89.

Keywords: Financial Innovation, Investment Banking, Underwriting, First-Mover Advantages, Demand Estimation.

The last twenty years have witnessed the introduction of a remarkable number of innovations in corporate securities.1 Most of these have been brought about by investment banks through the business of underwriting new corporate issues. It is also remarkable that investment banks have found it protable to develop new securities even when their competitors have been able to imitate them almost immediately and at signicantly smaller development costs. The empirical evidence so far has suggested that, despite these disadvantages, innovators are compensated with the largest share of the underwriting market. In this paper I estimate the demand of rms for the underwriting services of investment banks that use innovative corporate products. This will allow me to measure the dierent value to rms from raising money using a security engineered by an innovator or an imitator and thus explain part of the innovators market share advantage. The dynamic setup of the econometric model will allow a characterization of this advantage over time and shed light on the question of what makes innovators have a demand advantage over imitators. Product innovation in nance is particularly interesting because innovators face many disadvantages. Tufano (1989) provides evidence showing that imitation occurs shortly after the rst issue of a new security, leaving virtually no time for innovators to be the sole underwriters for that new product. He also observes that the development cost is signicantly smaller for imitators than for innovators. Further, the design of new securities is rapidly revealed to competing banks because of SEC rules of disclosure. Most strikingly, imitation cannot be precluded by any form of legal protection, e.g., patents. In his seminal contribution, Tufano (1989) observes that what compensates innovators for these disadvantages is the largest share of the market for underwriting corporate new issues: given a security, the bank that creates it is able to underwrite more capital than its largest imitator over all the securitys history. Why innovators are able to have such an unchallenged lead in these markets is perhaps the rst question that this evidence raises. This paper takes a rst step towards answering the question empirically. For that purpose I model the choice of a rm that needs to raise money externally through the issue of a security. This rm has to choose the type of security to be used and the bank that will underwrite it. Aggregating the choices of all the rms that need to make an oering, the model predicts the market shares of underwriting by dierent banks using dierent securities, conditional, among other things, on the characteristics of the banks (e.g., if they are imitators or innovators). Thus, after estimating this model, we can test whether innovators have

an advantage because their security is more valuable to the issuers. This paper introduces two features that allow us to get a better description of the facts of nancial innovation. One of them is the use of a framework of dierentiated products to model and estimate the demand for underwriting services. An inspection of recent innovations in corporate products suggests clearly that dierent securities are created to target dierent types of issuers or investors. For example, two similar debt products that tie the repayment of the principal to the performance of other indices provide dierent hedging devices to investors: the Stock Market Annual Reset Term Notes (SMARTs) are corporate bonds that pay a capped oating rate that is tied to the American Stock Exchange Oil Stocks Index while the Currency Protected Notes (CPNs) are bonds that pay a oating rate that is inversely tied to the Canadian six-month bankers acceptance rate. By taking into account the location of securities in a product space, it is possible to identify consistently a demand function for underwriting that depends on the price of underwriting (the underwriting spread).2 This is possible because we can associate the variation in market shares with the variation in underwriting spreads of varieties of the same security by dierent banks, and the variation in underwriting spreads of similar varieties which can be close substitutes.3 The other feature is that this study focuses also on the dynamic pattern of market shares. Instead of comparing the market shares of innovators and imitators over the whole history of a given innovation, I observe them over time and estimate the innovators demand advantage accordingly. This will reveal whether the innovators advantage is preserved steadily through the life cycle of the security or if imitators catch up with (or continue to fall behind) innovators. This dynamic setup also allows a comparison between the demands for sequences of securities. In fact, most securities appear sequentially, later ones as improvements of earlier ones. It appears that the life cycle of a security ends when an innovation that modies the older design is introduced. The empirical nance literature has not yet addressed extensively the question of why innovators have advantages over imitators. In fact, most authors have examined extensively the causes of the demand for innovative securities. The focus has been on trying to explain what made each particular innovation attractive to investors or issuers but not on why it is privately protable to develop such instruments. Miller (1986), for example, argues that the ma jor impulse to nancial innovations between the sixties and the

eighties came from ever changing tax codes and regulations that brought about prot opportunities (e.g., tax money saved) through the design of new nancial products that circumvented these laws.4 Not much work has been done, though, to try to solve the puzzle of why an unpatentable innovation is worth its R&D expenditure if imitation is less costly and immediate. In particular, not much has been said about what gives innovators an advantage over imitators. On one hand, some authors have tried to explain Tufanos stylized facts at the theoretical level by arguing that innovators are infra-marginal competitors, i.e., that have lower marginal costs than imitators. By moving rst, innovators may face lower search costs of identifying potential issuers and investors (Allen and Gale, 1994, Chapter 4) or lower average marketing costs if there is lumpiness in the set up costs of marketing networks (Ross, 1989) or if innovation signals skill and creativity credibly (Tufano, 1989). On the other hand, Battacharyya and Nanda (2000) provide a model in which the innovator is able to appropriate the value of its innovation and prot from it despite being imitated if it is costly for its clients to switch to other underwriters.5 ; 6 By contrast with these views, I analyze the possibility that the asymmetry between innovators and imitators is on the information owned by these two types of banks about the product. If some of the information that innovators have about the security remains private, a larger proportion of the R&D value added can be appropriated. In other words, the innovator can prot because it is imitated imperfectly. This possibility was explored in a theoretical paper by Herrera and Schroth (2000). In it, innovators of derivatives that move one period in advance receive private informative signals from their clients or the market. This allows them to oer deals that are more attractive to rms than what imitators can oer. For recent innovations in corporate securities, it is possible that imitation is imperfect. Equity-Linked securities and other derivative corporate products are sophisticated securities, specied by many parameters, some of which vary from deal to deal. Thus, it is possible that imitators cannot reverse-engineer perfectly the observed new securities from only a few deals. For example, the Equity-Linked Note (ELK) was the rst debt product to tie the repayment of principal to the stock price of another publicly traded company. The optimal choice of which stock to tie the notes to is observable but the knowledge of what stocks are optimal for dierent issuers or investors is a private component of R&D. Imitators that want to underwrite issues of ELKs for a potential client may know how to structure such instruments, but may not know exactly, from

what is observable, what stock to choose to tie the repayment of the debt of their client. Using data on all the new issues of equity-linked and corporate derivative securities, the (qualitative) results of the estimation of the demand for underwriting services can be summarized as follows: the demand for underwriting services using this type of securities is sensitive to the underwriting spread (i.e., its price); on average, the issuers demand for the underwriting service by an innovator using its own variety of the security is bigger than the demand for an underwriter that uses an imitative variety; this dierence disappears during the securitys life cycle, so imitators catch up with innovators; imitators catch up with innovators faster if securities are later improvements on past innovations. Thus, this paper provides the rst empirical test that the advantage to innovators may come from a bigger demand for the innovators products. While the empirical work directly shows how the innovators advantage evolves over time, the particular dynamic patterns found point strongly to the fact that innovators have superior knowledge on the products engineering and hence provide more value to the issuers. In fact, the dynamic patterns of competition it identies are consistent with the predictions of Van Horne (1985) when imitators enter the market.7 This paper also looks into the question of why and when are securities imitated, if so. I propose a simple model where imitation has the benet of providing some prots in case attempts at innovation of future generations fail. Imitation is also increasingly protable over generations as the demand for underwriting services with imitative varieties converges faster to the demand for the innovator. Yet if outsiders can also learn from observing the imitators complete deals, i.e., imitation spills-over information to non imitators, then imitation can become less frequent along the sequence of new securities. The evidence supports this view: regressions of the probability that a given security is imitated on the generation number of the security show that imitation is less frequent in late generations. In the next section of the paper I describe the data set I use and present some preliminary results that motivate the assumptions of the demand model I estimate. In Section 2 I develop the model that will allow me to characterize the decision of rms about the dierent types of underwriters (innovators and imitators). 5

First I present the formal setup for the discrete-choice decision problem of rms. I show that rms should derive additional value if they chose an innovator rather than imitators as their underwriter. Then I explain how this conjecture will be tested using multinomial logit and nested logit models of demand. In Section 3 I present the results of the estimation and Section 4 tests the robustness of these results to basically dierent denitions of what can be considered as innovative or as imitative securities. Section 5 derives a simple rule that rival banks follow to decide to imitate a new security or not. This decision depends on which generation is the security and how much imitation there has been in earlier generation securities. The predictions are then test using probabilistic regressions. Section 6 summarizes the main claims of the paper.

Some Preliminary Evidence

The data used in this research is obtained from the Securities Data Companys on-line databases of nancial transactions. I use all the private and public oerings of equity-linked and derivative corporate securities in the New Issues database and record as many details of the oer as possible: the name of the issuer, the principal issued, the name of the underwriter, the underwriters fees (underwriting spread), and details of the security, like oered yield to maturity, average life, spread of coupon over treasury notes, call options, etc. I merge this data set with the quarterly COMPUSTAT database (using six digit CUSIP numbers) to have nancial information about the issuer. In his study, Tufano (1989) uses all types of securities between 1974 and 1986. Here I restrict the sample to equity-linked securities because this type ts better the motivation that rst-mover advantages come from information asymmetries between underwriters. This type of securities are complex and underwriters have to choose many parameters to engineer such deals (In Table 1 I show the relative size of this market). Variations on mortgage backed products, for example, may be already too familiar to investment banks to hide something in their structure to imitators. The rst thing to realize about Equity-Linked and Derivative Securities is that they can be classied into groups. The SDC database identies 50 dierent types of them but a closer look indicates that some of these securities are related to each other in terms of their structure and purpose. For example, MIPS and TOPRS are instruments used by issuers to deduct debt interest payments from their taxable income, but the 6

former are issued by a limited liability company while the latter by a specially conformed business trust.8 To classify all the 50 corporate products in the SDC database I have relied on the experts opinion about the uses of these products for corporations and investors.9 I found ten distinct categories, which I summarize in Table 2. For the rest of the paper I will refer to these categories as groups or families of securities interchangeably. I will refer to each one of these securities as an innovation, since for each one there is a unique feature that distinguishes it from everything that already existed. However, depending on the group they are in and the order in which they appeared, I will assign to each security a generation number. For example, since MIPS were the rst tax advantage preferred note, I will call them the rst generation of this family, and TOPRS the fourth. I follow Tufanos (1989) criterion to dene the innovator of a security as the underwriter of its rst oer. Similar to what Tufano found, for equity-linked corporate securities we do observe that innovators have an edge over imitators in terms of market shares. 18 of the 50 products are imitated. In 13 of these, innovators lead in principal underwritten, and in 15 they lead in number of new issues. Also, imitation was fast: for 10 of these securities, the second underwriting deal was made by an imitator (see Table 3). In this paper I want to study why innovators have a competitive advantage over imitators. In particular, I want to test if issuers have stronger preferences for innovators than imitators as the underwriter of the oer. To choose an appropriate framework to study the demand for the underwriter of the issue it is worth examining the dynamics of the market for a new issue within each security and within each family. In Figures 1 and 2 I show the total quarterly principal underwritten by investment banks using the most important securities of two families. As we can see, each security seems to be popular for a certain period of time until a next generation appears and leads the market for issues of that group. Another interesting feature for some families is that the market share advantage of innovators over imitators seems to be bigger in the early generations. In gures 3 through 5 I show the evolution in time of the accumulated underwritten principal using a given security. Each gure represents a generation of the same family (convertible preferred stocks). For later generations, imitators end up accumulating a larger principal relative to the innovator. In some cases the innovator is overtaken. A similar feature is observed

in the family of index-tied principal appreciation securities (gures 6 and 7). It is less clear, though, if this is true for other families, like the tax-advantage debt or equity products. On average, still it seems the innovatorss advantage is weaker on late generations: Table 4 shows the average ratios of principal underwritten by imitators relative to principal underwritten by innovators. The ratio for rst generation securities is half the ratio that includes all imitated securities. The evidence above suggests that innovators seem to have an advantage over imitators that is stronger for earlier generations of securities within the family. Further inspection suggests that the appropriate framework to analyze the securities in this sample may be one of dierentiated products: it is clear from the denition of the 50 securities that these corporate pro ducts oer dierent benets to issuers or investors. Some provide a tax-advantage, others provide a hedge against the risk of defaulting on debt. Interestingly, within imitated securities there seems to be dierentiation across underwriters. For example, the withinvariety variation for some of these characteristics is smaller for imitators than for the innovator (Table 5). The price of underwriting, i.e., the underwriting spread, does not seem to dier signicantly, although the within security sample mean of the underwriting fee for innovators is larger than for imitators (Table 6). I will interpret this evidence as I present the econometric model and as I discuss the results of the econometric estimation. What I conclude from this evidence is that a useful model to describe competition between investment banks to underwrite corporate issues using this type of securities must be one of product dierentiation in an oligopolistic industry. Dierentiation occurs at the underwriters level, where innovators are distinguished from their imitators. Thus, from now on I will call a variety a distinct combination of a security and an underwriter.

2
2.1

The Demand Model

The Theoretical Approach

The model I present below is built to illustrate the decision making process of rms that want to raise capital and have to choose an appropriate security and the best underwriter, that is, the investment bank or bookmanager that will engineer the security and sell it. I will use a stylized model of the issuers preferences and

the underwriters information to derive the demand for underwriting services by the dierent banks using their variety of the security. The ob jective is to motivate a reduced form that can then be taken to the data and will allow us to test if those underwriters that invent the security have an advantage over the other underwriters (the imitators) that engineer the same security for their potential clients. Since the underwriter has to make engineering choices for the issue, an underwriter who is better informed on the choices he can make will provide the issuer with a larger value. I will show that the identity of the underwriter summarizes the quality of the engineering choices and thus the estimation of a demand function that depend on the banks characteristics will identify the bankers superior knowledge to underwrite that security. It is worth to point out that the ultimate source of the innovators advantage, the information asymmetry, is taken as given in this paper. The model that follows illustrates how the asymmetry is built into the demand function and what would be the empirical implications.10

2.1.1

The Setup

Firms that want to raise capital, the issuers, demand underwriting services from investment banks. These banks compete to underwrite the issue of a corporate security, and for this they oer dierentiated products: debt or equity types of deals that oer investors dierent payo schedules, horizons, call options, convertibility possibilities, etc. Thus, they compete for issuers that could use a non-standard variety of nancial instruments. A rm that needs to issue a security to raise capital is indexed by i 2 I : At a given period, there is a set of varieties of instruments, J = f1 ; 2; :::; J g from which it can choose one. Let g be an index for groups of varieties, g = 1 ; :::; G; such that G J and let there be a partition G = fJ1 ; J2 ; :::; JG g of J so that each set Jg contains those varieties which are closer to each other in terms of their characteristics. In this setup, for example, if consumers were choosing models and brands of a car then a set Jg would contain all brands of, say, Sport Utility Vehicles and some other set, Jh; would contain all brands of compact models. In our case, the groups in G are securities that have the same name, e.g., PERCS, LYONS or TOPRS and each variety would be determined by the name of the bank that underwrote the issue, e.g., PERCS by Morgan Stanley or PERCS by Merrill Lynch.11 Let b represent an underwriter in the set of banks, B .

The rm chooses one security and one underwriter among those that oer that security. A variety j 2 J is given by a unique (b; g) combination. Let uij be the value to rm i of choosing the variety, j: The setup for this decision is illustrated in Figure 8. The value function uij is central to this paper since it is the function whose parameters I will estimate using the data. I will characterize this function when I derive the preferences of issuers for underwriters that are innovators and underwriters that are imitators. The empirical literature that deals with the estimation of preference parameters in models of discrete choice uses special cases of the general specications of linear preferences by Caplin and Nalebu (1991) or Anderson et. al. (1989). In these studies, agents value a product according to a weighted sum of its components. These components, in general, are functions of the observable characteristics of the product. We shall see, below, how this structure is particularly appropriate for this study.

2.1.2

Preferences of Firms

In a typical underwriting deal of equity-linked or derivative corporate securities the rm will issue a security engineered by the underwriter. The rm has preferences over the set of possible structures for that security. An underwriting deal is dened by a combination of a vector of characteristics and an underwriting fee. If 2 is this vector, and p the fee, then an underwriting deal between a bank b and a rm i is fully characterized by f j; i; ; pg : The vector of characteristics could include, for example, the premium over dividend paid by common stock, the date of maturity of the security, the number of periods this security is protected from a call option by the issuer, or, more broadly, discrete variables that determine whether the security is convertible or not, if it is convertible to common stock or debt, etc. Let us start with a general random value function for the ith rm in a given period, t. If a rm chooses some variety j its value depends on how the security has been engineered, i.e., on j ; and its net income after the fee is paid. Let this value function be denoted by

u ijt = u (yit p jt; jt )+"ijt :

(1)

10

As shown by McFadden (1981), this function is continuous in y p and and twice continuously dierentiable in the same arguments provided that the function is continuous and twice continuously dierentiable in other goods and in : The term "ijt is assumed to be an additive random component that captures the random preferences of a given rm for a particular variety. It is unobservable to underwriters, and distributed independently with a continuously dierentiable cumulative distribution function, G(" ):

2.1.3

Underwriters

At each period underwriters engineer and price their own varieties of securities such that they maximize their prots. I assume that underwriters are Bertrand-Nash competitors in prices and . Their prots are given by

jt = p jtqjt c(qjt) ;

(2)

where the demand for the particular variety j is Z

qjt = q (pt ; t) = M t

dG (");
Aj( u)

(3)

and c( :) is the total cost, such that c0 (:) 0 : The demand for a particular variety is a function of the prices of all varieties the vector ; and the observable economy-wide shocks. The set Ajt (u) is the set of all the possible realizations of the individual shocks, (" i1 t; :::; "iJ t ); such that u ijt > uikt; k 6= j; i.e., the set of all the states that lead an issuer to a choice of variety j: M is a measure of the total number of rms, so that qj is obtained by multiplying M by the share of rms that choose j of the total number of rms that want to raise capital. The next lemma will simplify our work signicantly. It shows that an underwriters choice of jt that maximizes each rms individual utility conditional on its available information also maximizes the underwriters prots. I assume that u (:) is twice continuously dierentiable, strictly increasing and strictly concave.
Lemma 1 For a given price p jt; if jtmaximizes uijt (yit p jt; jt ;" ijt ; : ) then jt maximizes jt :

11

I prove this lemma in the appendix, but the intuition is very simple. Since the unobservable component of utility is independent of the characteristics of the security then the aggregate demand for a variety j is strictly increasing in the utility of any issuer. I have derived this result now because it will allow me to eliminate jt from the issuers value function, by substituting the optimal choice of jt : The objective in the next section is to explain how that choice diers across underwriters with dierent information sets and, as a consequence, how the value of rms diers depending on the chosen underwriter.12

2.1.4

Asymmetric Information

Suppose that the innovator has superior information about the engineering choices he can make for the potential issuers, given a particular security design. To formalize this idea, let 0 be the prior (common) knowledge set of all the possible characteristics that a security can have, and u ijt the utility function in (1), which is also common knowledge to all underwriters at some point in time. To relate this abstraction to the case of equity-linked or derivative securities, imagine 0 as a set of all possible engineering choices for a security before convertible debt was invented. Before the rst innovation, debt securities would be zero coupon or would have paid a xed or a oating rate, so any vector in 0 would have zero entries for other characteristics yet to be used, e.g., for convertibility to common stock. If an underwriter spends resources on R&D to come up with a new security, it will discover new possible combinations of characteristics a rm may nd valuable, possibly changing zero entries to add new dimensions to the structure of a security. The PERC, for example, was the rst issue of preferred stock convertible to common stock with capped and oored appreciation.13 Denition 1 An innovation is a new security design g; tied to the discovery of an engineering set g such that g 0 6= f g: When the rst underwriting deal is made using a new security the innovator reveals the new security design and a particular vector : The whole set of possible engineering choices for this particular security, g is kept private, though. The goal now is to show that the optimal choices of for innovators and imitators will dier, in general, and this will be reected in the value function of a rm signing an underwriting deal with either. Lemma 1 allows me to eliminate the vector of the said function and express the indirect utility 12

as a function of the banks identity. To see this, note that what distinguishes an innovator from an imitator is the set from which it can choose any : For the innovator, this set is 0 [ g ; and for the imitator this set is 0 [ k : Thus, we can summarize the identity of an investment banks by b = 1 if the banker is the innovator of the given security, and b = 0 if not. Lemma 2 Let v (y i p; b ) max u (y i p; ): For any couple of varieties j; l of the same security g; if j is

2 b

issued by an innovator then v j (y i p; 1) vl ( yi p; 0).

0 The result is trivial. 1 g is the set of choices of the innovator and g is the set of choices of the 0 representative imitator. Since 1 g g ; the result follows.

The Lemma above has established that given prices, innovators have an advantage over imitators. This advantage can be measured by the additional value to issuers if they were to choose an innovators variety. Let this dierence be named

vj v (yi p; 1) v (yi p; 0):

(4)

Note that an underwriting deal can be dened in a simpler way: now it can be summarized by f j; i; b; p g: This result is convenient for the estimation because what I want to capture is exactly the dierence in preferences for the dierent underwriters. Also, working with a value function, in which has been eliminated, avoids losing a large proportion of observations for which the full is not available. In other words, my interest is to distinguish preferences for these two types of banks more than to estimate the marginal valuation (and the derived elasticities) for a given characteristic of a security, e.g., years of call protection, yield advantage, etc. Another reason is that preferences for the choices of each attribute in may be complicated functions that make the estimation dicult. Thus, using a simpler function that summarizes all the attributes seems reasonable. This approach has been used by Caplin and Nalebu (1991), who use a utility function that is linear in functions that map the dimensions of the product characteristics onto a dierent space. Using their own example, the benets of a car could include only comfort and speed, but these could be more complicated functions of the physical attributes of the car. 13

2.1.5

The Demand Function

Here I discuss what Lemma 2 implies for the equilibrium in the market for underwriting. In this type of setup with dierentiated products, there is a demand function for each variety. The next proposition establishes that, ceteris paribus, the demand for an innovators variety of a given security is bigger than the demand for an imitators variety of the same product. Proposition 1 Lemma 2 implies that for two varieties j; k of the same security g and for a xed vector of underwriting spreads, p 2 RJ + such that p j = p k then

qj (p ; :) qk (p ; :)

if j is the innovators variety. I prove this proposition in the appendix. Note that if vj > 0 then the proposition above holds with a strict inequality. I will not show formally that in equilibrium innovators have bigger within-securities market shares than imitators. In fact, it is not obvious that this will be the equilibrium outcome. It is true though that, under mild regularity conditions on G (:) and v (:), the game becomes one of strategic complementarity.14 Moreover, if the best response function of the innovator shifts right if his advantage is positive, i.e., if v > 0; and if marginal costs are the same among underwriters then in equilibrium the innovator will charge a larger underwriting spread for his variety of the same security higher and have a larger market share within that security. If this advantage eventually decreases, then the innovators equilibrium price should decrease and converge to a symmetric equilibrium as the advantage goes to zero.1 5

2.1.6

The Demand Advantage through time

Proposition 1 has established that, if part of the set of engineering choices the innovator can make to engineer a security is kept private, then the demand for his variety will be larger than the demand for any imitators variety, that cannot be engineered as well as the innovators. Clearly, though, as the innovator completes more deals, more possibilities for engineering are disclosed to potential imitators. With time, the set from 14

which the imitator picks becomes larger, and eventually gets arbitrarily close to the innovators set. In fact, in this setup the imitator only learns the innovators set element by element through observations, and it may be the case that the imitator can reverse engineer this set much faster, from the observation of the innovators choices and some prior knowledge of the issuers preferences. In any case, it is clear that with time the imitators choice set gets larger. Lemma 3 If v j is decreasing in time, then the demand for the imitators variety converges monotonically to the demand for the innovators variety. As a corollary, if next generation innovations are marginally decreasing improvements of the innovators then convergence should be faster for later generations. Namely, if the initial innovators advantage v j gets smaller as new products are introduced, we would expect the advantage to be shorter lived, which would show up empirically by a demand function for the imitative varieties that, ceteris paribus, converges more rapidly to the innovators.

2.2

The Econometric Model

I argued above that the market of underwriting new issues using equity-linked securities and other corporate derivatives may be well approached as one of dierentiated products. Each variety oered is given by a combination of a security structure or name and the identity of an underwriter. In this section I present the model that I take to the data. This model will be a reduced form equation obtained from aggregating the individual rms demands for the dierent varieties. I establish dierent sets of assumptions for the aggregation of individual rm demands, and the results will be dierent reduced forms to estimate: the multinomial logit and the nested logit demand models, each one requiring a dierent method of estimation.

2.2.1

The General Setup

I consider each time period t = 1; :::; T a dierent market in which an issuer i chooses its desired variety j: As the standard of the empirical literature of discrete models of demand, I will specify the value of this issuer as a function of observed and unobserved characteristics of the issuer and of the product oered by

15

the underwriter, and of the relevant parameters. Let

u ijt = (y it pjt ) + xjt + jt + "ijt:

(5)

The rst term on the right are the net proceeds of the issue, which are assumed to be separable and linear.16 The second term captures the indicators that distinguish an innovator from an imitator. The vector xjt is then an index of v j that will include all the variables that reect dierences in the information owned by underwriters. As we shall see below, I will not only employ dummy variables that capture the preference rms have, on average for investment banks with superior information. I will also try to identify the dynamics of this advantage by interacting the identity of banks with the number of time periods after the security has been imitated, and the order in which the security appears within its family. I will also account for the fact that banks may acquire reputation as experienced underwriters of this type of securities based on their history as innovators within a family or any other type of equity-linked or derivative corporate security. The value of choosing alternative j to raise capital can be decomposed into its mean, jt = pjt + xjt + jt and a deviation from it, y it + "ijt : The unobservable (to the econometrician) characteristics of the security j itself are captured by j ; while the deviation term is used to account for the heterogeneity of rms preferences. "ijt would be a purely idiosyncratic, mean zero, shock. Below I will explain briey the dierent ways I will estimate the parameters of the value function.

2.2.2

Logit Demand

Berry (1994) shows that if we under some distributional assumptions about "ijt we can identify the parameters of (5) using the observed market shares of all varieties, j: In particular, assuming that "ijt has a density function f (" ) = exp( exp(")) the market shares predicted by the model, b sjt ; which are obtained by integrating all the realizations of unobservables that lead to a choice of j over all other varieties, are given

16

by the well-known closed form solution (the logit formula):

s bjt( t ) =

1+

ejt PJ

l=1

elt

(6)

Note that the average utility of the outside alternative is normalized to zero (0 t = 0) and that the term yit drops out because it is common to all the choices. The logit formula has the property that the market shares are uniquely pinned down by the average utility of a choice j: Thus, and can be obtained from a regression of the dierence of the logarithms of the observed market share of j and an outside alternative, on x jt and p jt ; e.g.:

ln sjt ln s0 t = p jt + xjt + jt :

(7)

The estimation of this model is simple. The challenge is to nd appropriate instruments for the price because it is very likely that it is correlated with the unobservables, jt : This is a typical problem found in studies that use product characteristics as regressors. In this case, xjt uses issuers characteristics that summarize the full description of the product, so it is less likely that jt contains product unobservables correlated with the price. There may still be other costs of imposing this particularly convenient structure. As a consequence of assuming that the "ijt are independent and homoskedastic, the odds ratios of choosing one variety over another do not depend on the value of other varieties. This can be problematic: suppose the demand for MIPS were evenly split between Goldman, Sachs and Merrill Lynch, and each were half the demand for Salomon Brothers ELKS. This model would predict that the ratio of the market share of Goldmans MIPS to Salomons ELKS would still be one half, even if Merrill increases its underwriting fee by any magnitude. Implicitly, the business lost by Merrill Lynch would be absorbed by both Goldman and Salomon so as to preserve the ratio, never mind that Morgan Stanley and Merrill Lynch oer close varieties and Salomon oers a dierent product.

17

2.2.3

Nested Logit

The nested logit allows a richer pattern of substitution than the simple logit and at a small additional computational cost. The decision to choose a variety is now represented by a tree as in Figure 8. The preferences of rms are allowed to be correlated within groups. In this case, dierent varieties of the same securities oered by dierent investment banks would be closer substitutes of each other than any other security. In this case, the utility of a given choice, j; can be modelled as a restricted version of (5), allowing for random coecients, ig ; on security specic dummies. Thus, we have X
g

u ijt = jt +

djg ig + (1 ) "ijt ;

(8)

where djg = 1 if j 2 Jg ; and "ijt is still assumed to be independently and identically drawn from a Weibull distribution. If the random coecients are also assumed to be drawn from a Weibull distribution, then so is the term +(1 )": The degree of within group correlation is given by : if it approaches one then so does the within security correlation of utility levels, and if it approaches zero then there is no within security correlation and we are back to the logit model. Due to this assumption there is an analytical solution for the predicted market shares of underwriters within the security:

sj=gt ( t; ) b Dg t

= =

h2 Jg

ej t=1 ; Dg t X eht=1 :

(9) (10)

The overall market share is

sjt( t ; ) = b

ej t= 1 : P 1 Dg t[ f 2G Df t ]

(11)

Normalizing 0 = 0 ; which implies D0 = 1 then ; and can be recovered from an Instrumental Variables regression of the dierence of the logs of the observed market share of j and the outside alternative 18

on x jt ; pjt and s j=gt because

ln sjt ln s0 t = ln sj=g t p jt + xjt + jt :

(12)

Again, instruments for prices and additional instruments for within security market shares must be used to obtain consistent estimates because both variables are endogenous.1 7 ,1 8

2.2.4

Issuer Heterogeneity

To enrich the substitution patterns in the demand model I incorporate data from COMPUSTAT about all the rms that did new issues using equity-linked security in the sample. This will introduce heterogeneity that will make the cross-price elasticity depend on the issuers characteristics. In this case, I consider a vector of f rm characteristics f t ; each one to be interacted with the price to obtain the following estimable relationship:

ln s jt ln s0t = pjt + xjt + p jt ft + jt:

(13)

As we shall see in the results, the cost is that we will lose a signicant proportion of the observations in the sample. Many of the issuers of equity-linked securities had no record in COMPUSTAT.

2.3

The Data

As I mentioned before, the SDC Database of New Issues records all the public and some private oerings made since 1962. For securities dened in SDC as equity-linked or derivative corporate securities there are 662 oerings up to March 2001 (the rst issue, a LYON, was made in April of 1985). There are 50 securities and a total of 98 varieties. I compute the varieties market shares over the whole market of new issues and over the varieties within the security at dierent time periods. I divide the whole sample in time periods rather than aggregate the data by varieties over the whole time span studied. Overall aggregation would reduce signicantly the number of observations (to 98) and would also eliminate the time variation of market shares and underwriting fees, compromising seriously the

19

consistency of the estimators. Thus, I treat each time period as an independent market, so that there is a demand function for each variety at each time. The parameters of this function are identied by crosssectional variation in prices, in the identity of the underwriter, and the issuers characteristics and by the time variation in prices and issuer experience. The panel structure of the data is crucial since I want to study the dynamics of the advantage to innovators. To form the panel I must choose the length of each time period, though. The shorter the length of each period increases the size of the usable data set but increases the risk of aggregating very few or unique deals per period, which would increase dramatically the variation in the market shares. To avoid arbitrariness in the choice of the length I do the estimations at four dierent levels of aggregation: using 16 periods (annually), 8 periods (biannually), 11 periods (18 months) and 12 periods (16 months). In this way we can also have an assessment of the robustness of the results to this choice. The panel is unbalanced because not all securities are oered at each period. Only two varieties are oered in the rst period and 98 in the last. I consider standard equity as the outside option to issuers, i.e., standard equity is the variety j = 0: I approximate the total size of the market for new issues using

M = q0 + q1 + ::: + qJ :

(14)

The unit of demand is number of deals, not dollars underwritten. This assumes that rms set ex-ante the amount of cash they want to raise in the oer, and the choice I model here is the choice of the security and the underwriter.

2.3.1

Variables Overall market shares, s j ; are the observed aggregate number of deals for that variety in

Market Shares

a given period divided by the total number of new issues. Within-security market shares divide the number

20

of deals by the total number of issues using the relevant security:19

sjt sj=g t

= =

qjt ; Mt q P jt

l2J g

qlt

Prices

Prices of underwriting are the fees charged by the investment bank that leads the syndicate of

book managers of an oer. They are expressed as a percentage of the principal underwritten and called underwriting spreads. Usually this spread can be disaggregated in the underwriting fees and management fees. This disaggregation is seldom observable though, so the price variable I will use is the total spread.20

Demand Shifters in xjt

The demand shifters that do not capture the information asymmetry between

underwriters are variables about the underwriters experience and reputation issuing this type of securities. I use the total number of innovations in equity-linked products and innovations within the particular family of the security accumulated by the underwriter. I use time period dummies to control for observable economywide shocks and group dummy variables.

Advantage to Innovators

One way to test if innovators have advantages on the revenue side is by

including a dummy variable that equals one when the underwriter was the rst to issue that security. A positive estimate of the coecient of this variable would imply that, on average, rms have stronger preferences for innovators. In the model presented above, the innovator has an advantage because it holds private information about the security issued. However, this advantage could diminish as more deals are completed by imitators. Thus, we would expect the estimate of the coecient of the innovator dummy interacted with the number of deals after the security was imitated to be negative. Moreover, if the security is a late generation of a given group, more information about this type of securities would have been aggregated, and we would expect imitators to learn the innovators private information faster. Thus, I also interact this dummy with the generation number to get a richer characterization of the dynamics of learning by doing.

21

Formally, I model these dynamics by specifying the component vj of the rms utility as:

v jt = 0 i j + 1 i j genj + 2i j genj e t;

(15)

where the dummy variable i j = 1 if the variety j is the innovators variety, gen is the generation number of the security and e t the number of time periods since the rst imitation. Issuers Data I use nancial data from COMPUSTATs quarterly database that matches the period of the

oer. I use the total market capitalization to measure the size of the rm. I also use indicators of common equity, preferred equity, short term, long term debt and subordinated debt all expressed as percentages of capitalization.

2.3.2

Instruments

Since it is very likely that the price is endogenous, instruments are needed to obtain consistent estimates of the parameters of the model. In the case of the nested logit specication, the within-securities market shares are used as a regressor and these are possibly endogenous too. To choose appropriate instruments I follow the suggestions of Berry, Levinsohn and Pakes (1995) and Berry (1994). Instruments for the underwriting spread (price) include the averages of characteristics of the security over the competing varieties, like years of call protection, years prior to call at par, percentage yield, which should not be correlated with the error term since the advantage term summarizes all characteristics of the security. By the same token instruments for the within-security market shares include characteristics of other underwriters in the same group (e.g., total and within family accumulated innovations by the other underwriters of the same security). To test if these instruments over-identify the parameters of the models I perform a Hausman test of over-identifying restrictions.

22

3
3.1

Results
Logit Demand

To serve as a benchmark, rst, I t the simplest, yet most restrictive, demand model: the multinomial logit. I only report here the results for the aggregation at 12 time periods for the sake of parsimony (the subsequent estimations will include all aggregation levels to show the robustness of the results). Table 7 reports the estimates of the parameters of (7), allowing for unobservable (to the econometrician) attributes in the dierent varieties and using an instrumental variables method to account for the correlation between the price and the unobservables (the standard errors were estimated using the Huber/White variance estimators, allowing for heteroskedasticity and serial correlation within securities). I t two models: one that only includes the innovator indicator from equation (15) (reported in the Base Case column) and another one that species the full dynamics in (15). I use time perio d-specic dummies and xed eects for the security group. We can see, for both columns, that most estimates have the expected sign. The underwriters fee (the price) is signicant at the 90% level. In both cases the estimated coecient is negative, which indicates that the market demand for a bankers variety is downward sloping in the fee. Note that I report the estimated coecient of the underwriting spread, so the estimated is positive. If, as assumed, the data of the underwriting markets were generated by a model of oligopoly then the elasticity of demand at the fees observed should be bigger than 1. In other words, bankers should be pricing their deals in the inelastic portion of the demand curve. In Table 7 I report the number of estimated demand curves (of 323) that violate this condition (i.e., that are elastic at the observed fee). Both logit models imply 50 and 49 inelastic demands, respectively, out of 323 estimates. The average elasticity, though, is well above 1. The average value to an issuer increases if the number of innovations in equity-linked securities accumulated by its chosen underwriter increases. The negative sign of the number of accumulated innovations within that securitys group, though, is an unexpected anomaly. In both columns, the innovator dummy has a positive coecient, signicant to the 99% level of condence, suggesting stronger preferences for the innovators variety, ceteris paribus. The second column reveals an interesting result. The coecients on the innovator dummy, on the dummy interacted with the generation

23

number, and on the dummy interacted with the generation number and the time after imitation are all signicant to the 99% level of condence. The estimate of the coecient of the rst interaction term, i gen; is positive, revealing that the later the generation, the higher the average initial advantage of the innovator. The second interaction term, i gen e t; has a negative estimated coecient, showing that this advantage faster the later the generation. Since I estimate this model using instrumental variables, I test if the restrictions imposed by using the chosen instruments over-identify the parameters of the model. I perform a 2 that tests jointly if the model is correctly specied and the instruments over-identify the variations in the endogenous variables. The test statistic and its p-value are also reported in Table 7. In any case, the null hypothesis of correct specication and over-identication is rejected. Rejection is inconclusive about the source of the problem (misspecication or lack of identication), but, as we mentioned before, the logit model is a simple version that imposes restrictive substitution patterns across dierent varieties of underwriting services. In fact, the previous results were obtained under the assumption that varieties of the same security were as close to each other in the product space as varieties of dierent securities. Group dummies may have accounted for proximity within the family, but not within the security. The results that follow are for the nested logit model, that deals with this problem.

decreases in the number of periods that imitators have been in the market, and that this advantage diminishes

3.2

Nested Logit Demand

The estimation procedure for the nested logit demand model is similar to the one used for the multinomial logit. The dierence is that, here, I include as a regressor the within-security market shares for each variety in order to obtain an estimate of the intra-security substitution eect. For this matter, additional instruments must be used since the new regressor is believed to be correlated with the varietys unobservable characteristics. This model was tted for the four dierent aggregations of data: 8,11,12, and 16 periods. The results are shown in Table 8. The estimated coecient of price still has the correct sign for all the aggregations. It is signicant at least at the 95% level but for the case where t = 8 (where its p-value is 0.121). The estimated elasticities

24

increases sharply after accounting for the substitution eect of the underwriting services of banks oering similar varieties. As a consequence the implied number of inelastic demands is much smaller (10 at most). The estimated coecient for the substitution parameter is signicant in all cases, and the estimate is within the appropriate bounds, 0 and 1 (0:618 < b < 0 :745 ). This result is consistent with the theoretical setup in which varieties within a security type are closer substitutes than varieties outside the security type: issuers switch bankers before using a dierent security structure. For all these cases I have tted the model that describes the dynamics of the innovators advantage. The estimates of the innovators advantage component reveal the same dynamic pattern as before: the innovator dummy has a positive estimated coecient, as well as the dummy interacted with the generation number. The coecient of the interactions of the innovator dummy with the generation number and the number of periods after the rst imitation is negative. Figures 9 to 12 illustrate better what the estimates for the dynamics of the advantage mean in terms of the time during which issuers value innovators varieties more than imitators. I plot the estimated advantage
In of the innovator, i.e., the predicted ratio of the innovators market share to the imitators ( s sIm ), in the

vertical axis against time, measured in years, in the horizontal axis. We can see that, for all aggregations, the innovator of a rst generation security has the smallest initial advantage over his imitators. In all cases, this advantage disappears slowly (in 12 years, on average). In sharp contrast, an innovator of, say, a seventh generation security enjoys a bigger average initial advantage over other competing underwriters. This advantage, though, will be gone shortly after the products second year of having been imitated (that happens when both market shares are predicted to be equal to each other). One possible interpretation of this result is that late generations are often very complicated modications of existing securities. At rst it is dicult for imitators to learn how innovators are engineering the deals, but in time they should learn faster given that more information has been aggregated about the security type or of the family of securities. The hypothesis that the model is over-identied and correctly specied is rejected in the aggregations over 11 and 12 time periods. Rejection may be due to the fact that the instruments chosen do not introduce sucient independent variation themselves to account for the variation in all the endogenous variables of the model (price and within-type market shares). It is also possible that the model is not fully specied and the

25

instruments themselves are correlated with other excluded exogenous variables. However, it is interesting that the model is over-identied when time periods cover 2 years (when t = 8). It is possible that within shorter intervals, the instruments used are strongly correlated between themselves, while this may not be the case for longer periods. It is also worth pointing out that, the rejection of over-identication at some levels of aggregation is not strong evidence against our choice of instruments, since almost all of the estimates are consistent across all the aggregations.

3.3

Issuers Heterogeneity

The estimation by instrumental variables of the logit and nested logit demand models above may have allowed us to obtain consistent estimates of the own-price elasticity, but it may still yield implausible crossprice elasticities for varieties in dierent groups. Also, the test of over-identifying restrictions for the nested logit specication revealed that the model may still have not been completely specied. In Table 9 I show the results after adding the characteristics of the issuer to the model via interactions with the price variable. Although it is not my goal to estimate these cross-price elasticities, adding heterogeneity will dierentiate own-price elasticities by the type of rms. Despite the loss of observations when using issuers data, Table 9 shows results that do not dier qualitatively to the previous ones. The same dynamic behavior of the innovators advantage is observed in all four cases. For all aggregations over time, the initial advantage is bigger than in the previous specication but it also decreases at a much faster rate. On average, for all cases of the model with interactions, the advantage of each generation would be gone almost by the time predicted in the nested logit model without interactions (see Figure 13 for the case when t = 8) : Of the ve issuers variables that I interact with price, only market capitalization and preferred stock as a percentage of market cap were found to be signicant at a level higher than 90%. Their estimated coecients were both positive. One possible explanation is that market capitalization is an approximation for the available sources of nance to the issuer. Similarly, since most of the varieties are forms of preferred stock or convertible to preferred stock, rms with a larger proportion of this type of stock have more available instruments to raise capital and thus their demands are more elastic to underwriting spreads.

26

Adding rm specic data has improved the t of the model in the sense that the estimated elasticities of demand are higher and the number of predicted inelastic demands has decreased. Moreover, with the additional rm specic regressors the null hypothesis of over-identication and correct specication cannot be rejected in any case, even with 90% condence. And as before, the estimates are consistent over all the estimations.

Robustness Analysis and Further tests

In this section I analyze the data set further to discuss the appropriateness of some of the assumptions made earlier and test empirically if the results presented above are robust to changes in some of these assumptions.

4.1

Trademarks on Securities Names

It is possible that some securities that have been considered here as innovations may only be replications of existing designs but marketed with dierent names. If innovators register a trademark for the original name of the security, its imitators would be confused as innovators of later generations. Thus, it could be incorrect to consider such varieties as innovative and their rst underwriters as innovators rather than imitators. An inspection of all the 50 securities in the data set identies dierences in the design of almost all securities with respect to their previous generations.2 1 Moreover, the setup of the estimation allows for intergenerational substitution eects. However, these eects are by construction smaller than the intragenerational substitution eect. Also, the fact remains that probably the aggregation used so far denes too many varieties as innovative. It is worth noting, though, that labeling more banks as innovators rather than imitators in later generations is likely to bias the innovators advantage estimate downwards because late generations have typically small market shares and are not as imitated as early generations. The current denition of an innovator associates observed small market shares to innovators, some of which could be imitators. Thus, since the possibly downward-biased estimate of the innovators demand advantage over its imitators is already positive and signicant, our inference from the current results should not change much qualitatively. To assess the robustness of the results I propose two alternative exercises. First, I estimate the same parameters having excluded from the data set all those securities that were not imitated and whose name 27

was trademarked. I searched for all the 50 security names in the United States Patent and Trademark Oce records to see if there is a trade mark for any. This criterion eliminated 7 securities, and 77 issues from the sample. The remaining securities were either imitated or could have been imitated. Table 10 shows the new estimates, which are very similar than the ones obtained before. The estimated coecient of price still has the correct sign for all the aggregations. The estimates of the price elasticity of underwriting demand are similar, and so is the estimated coecient for the substitution parameter. The estimate is still always between 0 and 1 (0 :654 < b < 0:704 ). The estimates of the innovators advantage index follow the same

pattern as all the previous results. Note that the estimated initial advantage of the innovators is slightly larger with respect to those in Table 8, but the estimated speed at which this advantage decreases has increased.

4.2

Underwriters placement ability

It is reasonable to expect that, all other things constant, the value to an issuer to choosing a bank with a higher likelihood of placing successfully the security should be higher. In the previous estimations we have used the number of accumulated innovations in the group and in the equity-linked class as proxies for the reputation of the underwriter in these markets. While these may be good proxies of the innovativeness of the bank, they may not necessarily capture well the potential of the bank for selling them. In Table 11 I show the estimates for the nested logit choice model, excluding trademarked securities, and adding as a regressor the cumulative principal in equity-linked securities underwritten by the bank of the particular variety until the time period prior to the observation. The estimates for the parameter associated to this variable are strongly signicant and positive for all the time period aggregations. This result suggests that issuers do derive more value from hiring a bank that has been able to underwrite a larger volume in the equity-linked class.2 2 ;2 3 The inclusion of this variable has not aected most of the previous results. The estimates of the innovators advantage have changed very little and are still strongly statistically signicant. With this specication, the underwriting demand is less elastic to the underwriting fee (the estimate is not signicantly dierent from zero in all time aggregations) and the elasticity of substitution between products is lower (although,

28

always signicantly dierent from zero and between the proper bounds). The goodness of t measures have increased also. These results imply that banks with good prospects to sell the oering of equity-linked securities successfully can actually retain their advantages for longer periods.

4.3

Product Groups as Observational Units

The second approach that I use to overcome the problem of identifying correctly the level of competition in this data is to recompute underwriter market shares, this time over product groups rather than over individual security names. I redene a variety as a banker-product group pair, and compute market shares accordingly in all time periods. This aggregation is the other extreme of the previous one: here, only the products that start a category are considered as innovations, while before, any security name was considered as such. Thus, this aggregation rules out the fact that some generations are actually innovations in their own right. Note that, with this aggregation, the generational eects cannot be assessed because all subsequent generations within a group are considered imitative varieties. A variety now is dened by a bank-group pair. The number of total varieties falls from 98 to 60, and so does the number of usable observations for the estimation. The results change little with this aggregation (see Table 12). In fact, this aggregation seems to t the data better: the joint hypothesis of correct specication and over-identication through the instruments are never rejected with 99% condence, while the R-squared coecients are higher. The same dynamic pattern concerning the innovators advantage is observed: with time it disappears. In this case, the size of the advantage is initially smaller, but this is not surprising given that this estimate measures the innovators advantage over a typical bank in the product group, rather than over a bank underwriting the same security only. For this aggregation, the cumulative equity-linked principal that the bank has underwritten the period before also increases the demand for its underwriting services within the category of securities, as so does the accumulated number of innovations. Controlling for these eects, the coecient of the underwriting fee is also signicantly dierent from zero, as well as the substitution parameter.

29

4.4

Further Comments

Another concern may be that securities underwriters actually cooperate, so the demand for a given banks variety of a security is not independent of the demand for other banks varieties, if these banks were part of the others underwriting syndicates. It is well known that underwriters frequently form syndicates to structure the deals and sell the securities. Moreover, as Nanda and Yun (1995) argue, comanagement of underwriting may be a form of cooperation that overcomes the free-rider problem in securities innovation: if all the competing banks underwrite the rst issue of an innovative security jointly, they can share the risks and the prots of the innovation in order to minimize the spillover of development costs. Thus, they argue, one would expect that underwriters would team up in syndicates more often when the market for the innovation is small, or when a potential innovator does not expect to capture a large share of the new market. However, there are strong reasons to believe that the joint management of deals is not common practice in the Equity-Linked and Derivatives class. Of the total 662 equity-linked deals observed, only 17 are co-managed. In 15 of these cases, one of the co-managers was indeed the innovator, but in only three cases the co-managed deal happened before the 5th deal (in fact only one rst deal is co-managed). We only observe two deals where the innovator shared the management with a bank that also underwrote any other deal using that security.2 4 ;2 5 Another important point worth noting is that the in the sample use one cannot observe when where varieties made available by imitators, i.e., since when were rivals competing. The econometrician can only compute imitators market shares after they have underwritten their rst issue. For example, if an imitator appears on period 5 there are no observations for this variety in period 4 or anytime before when the security was in the market but not yet imitated. There is the possibility that the bank was oering this variety well before we can observe its rst deal. The exclusion of these observations may bias the results, but the bias is to reduce the estimate of the innovators advantage. Since entry of imitators can only happen sooner than we can attribute it, the dataset would actually have some more zero market share observations associated with imitating banks. Thus, the estimate we have now, which is already positive and signicant, could even be larger. Note too that as we increase the size of the time horizon for the computation of market shares, the problem noted above loses its eect on the results. This is because for larger time periods, the imitator

30

appears most likely in the same time period as the innovator, and the potential for bias is smaller. Precisely, we observe that the estimated innovators advantage is largest when the size of the time period decreases (i.e., at t = 8 the estimate is highest). Following this discussion, a remaining concern is, why are some securities imitated and others not. I address this question in the next section.

Patterns of Imitation and Innovation in Equity-Linked Securities

The evidence discussed above has shown robustly that imitators are at a disadvantage with respect to securities innovators: the demand for their underwriting service is, ceteris paribus, smaller. The disadvantage is smaller though for later generation securities. In this section I use this nding to try to answer why some securities are imitated and others not. Understanding why and when these innovations are imitated sheds some more light into the nature of the game that investment banks play when underwriting issues of corporate derivatives.

5.1

The Decision to Imitate

As I pointed out earlier in the paper, there are 43 securities that could have been imitated, but only 18 of these actually were imitated by rival banks. Even if imitation were costly, and underwriting deals as an imitator rather than an innovator puts the bank at a disadvantage, it is clear that imitation should be less costly than innovation (e.g., free-riding the costs of development and legal advice). Table 13 shows some preliminary ndings that help to narrow our focus on the possible explanations for why some securities are not imitated. Panel A of the table shows that imitation seems to be more frequent in the rst generations than in the later ones: the null hypothesis that no imitation is associated with later generation securities can be rejected with 91% condence. In Panel B I group generations in three categories: rst generation securities, securities between generation 2 and 6, and later generation securities. In this case, the statistical association is stronger (the p-value is 0.047).26 Panel C suggests that banks are more likely to be the innovators of a

31

given generation if they had previously imitated or innovated in the same product group: we can reject that there is no association between having previously underwritten deals in the same product group and being a next innovator with 96.6% condence. Thus, it seems that doing deals helps banks to develop new products but that this helps less later on in the product sequence. Imitation is also increasingly protable, since the demand for an imitators variety increases faster for later generations. This seems to be in contradiction with the fact that there is less imitation later because, even if imitation is costly, if it becomes more attractive it should be more frequent. A possible conjecture is that the benets of imitation spill-over to rival underwriters that dont imitate: if they observe more imitation, then it is more likely that they will be successful at attempting next generations innovation. If spill-overs are large enough, banks will imitate less often. Probably a more obvious conjecture is that banks have dealt enough by the time they attempt at innovating late generation securities, so there is little gain from imitating if it cannot increase enough anymore the future success probabilities. But even in this case, imitation prots are higher due to the higher demand. Moreover, a bank can always deal its innovations and imitate other banks products simultaneously. To understand better the decision to imitate, consider the following very simple decision rule, whose predictions I will confront later to the data. A new security, which is the g th generation in a sequence of related innovations has been introduced and E banks (entrants) have to decide to imitate the product or not. Imitating product g increases the chances that the bank may develop a next generation product, from a probability of z( g) to z (g ) + z: The other benet of imitation is that, in case of failure at attempting an innovation, the bank can always enjoy imitators prots, Im (g ): The evidence so far shows that Im (g ) is increasing in g: Suppose that Im (:) is dierentiable. Imitation costs F , and if one bank imitates then this will increase the others probability from z (g )+ h z: The spill-over of imitation is thus hz; where we assume that 0 h 1 to rule out the case where a bank benets more from observing imitators rather than imitating itself: An attempt to innovate a next generation costs F >F and it pays innovators prots In if successful. Assume that all the banks have an interest in developing a next generation. Consider an equilibrium where every bank chooses the same probability, p; of imitating the current

32

generation. For any bank, if they are randomizing between imitating or not, it must be that the probability of imitation chosen by the remaining E 1 banks make the E th indierent. The payos from imitation are:

(z (g ) + z ) I n + (1 z (g ) z) Im (g ) F F :

(16)

If the remaining banks are imitating with probability p each, then a bank that does not imitate can benet from a spill-over to the probability of innovating successfully. The probability is 1 (1 p )E 1 : Thus, the payos to holding back from imitation are

[z (g ) + (1 (1 p )E 1) hz ] In F :

(17)

Note that it is crucial that imitation is costly, because otherwise, imitation would dominate and be observed always. The equilibrium probability of imitating a given generation g is a value for p in the unit interval that equates (16) with (16). From this equality, dierentiating with respect to g we obtain:

0 dp [1 z z( g)] 0 Im (g) z ( g) Im (g ) = : dg (E 1)(1 p )E 2h z In

Thus,

dp dg

< 0 if and only if z0 (g ) Im (g ) > [1 z z (g )] 0 Im (g ) for any value of g:

Since the probability that a generation g is imitated is 1 (1 p )E 1 ; the following predictions can be made: 1. If imitation does not spill-over to non-imitators then imitation should be independent of the generation number. 2. If imitation spills-over, and if z (g ) is increasing in g; and z0 (:) is high enough then imitation should be less frequent in later generations.

33

Note that

dp [1 h (1 (1 p)E 1 )] = > 0; d I n (E 1)(1 p) E 2 h In which implies that: 3. Imitation is more likely if the expected prots from innovating a next generation are higher. Note also that there may be heterogeneity in the players in terms of their experience imitating and innovating past generations. The next section describes how these are addressed empirically and how these predictions are tested.

5.2

The Evidence on Imitation

To test these predictions I t probabilistic regressions of a security being imitated or not on the generation number of the security. I exclude the securities whose names were not trademarked in the US Patent and Trademark Oce. This leaves 43 of them. As controls I use: the total number of underwriting deals accumulated by previous generations in the same product group, as a proxy for the expected size of the market of the next innovation; the number of previous generations in the same product group that were actually imitated, as an additional source that speeds up the spill-overs and makes imitation less desirable; dummy variables that indicate if any of the underwriters of the given security were also dealers of the rst generation, the second, and so on. Table 14 shows the estimates of the regressions. The covariance matrix has been estimated using Huber/Whites estimator, which is robust to correlation of the securities within the group and robust to hetersokedasticity. Columns 1 and 3 use the generation number as the main regressor, and Columns 2 and 4 use instead the same categorical variable used in Panel B of Table 13. Columns 1 and 3 show that a given generation is imitated with a probability that is smaller by 0.08 or 0.07 than the probability that previous generation is imitated, respectively. These estimates are signicantly dierent from zero with 99% 34

condence. Columns 2 and 4 we see that the probability of imitation is decreases for generations 2 and 3, and decreases further for the remaining generations. As expected, in larger product groups, the next product is imitated more frequently, and this eect is also statistically signicant to the 0.01 level in all cases. Note that the regressions control for the number of imitated previous generations, which is also statistically signicant and, as expected, reduces the probability that the next products in the sequence are imitated. Also, securities whose underwriters had also dealt the groups rst generations are imitated with a higher probability, all other things constant. In other words, securities that are imitated with a higher probability are those for which their innovators or imitators have had accumulated experience innovating or imitating the rst generations. Having experience in the second generations has a negative eect on imitation (experience in further generations was rejected of all specications). This result, though, may just be a reection of the main hypothesis: that banks have less incentives to imitate later due to the spill-overs of other imitators.

5.3

Further Comments

The results shown in this section and before have characterized to a large extent the nature of the game played by underwriters of corporate derivatives. In these markets, the largest, most reputed investment banks compete to underwrite issues which are, on average, much larger than common stock deals. To seize high prots these banks dierentiate from others using their innovative ability, their engineering skills and their placement capacity. Success at innovation gives them discretion to choose underwriting fees through a demand advantage. Since this advantage dissipates with time, it is not surprising that banks typically dont co-manage deals with others: co-management could disclose too much private information to potential imitators and speed up the elimination of innovators advantages. Banks also need to imitate their rivals to stay in the game at the early stages of development of innovations. Imitating early generations seems to increase the chances that banks will develop innovations in the future. But even as imitation becomes more protable along the innovation sequence, less imitation is observed later. Imitation becomes less frequent faster if there has been more of it in the earlier generations. This may be due to the fact that banks that dont imitate also learn by observing deals completed by rival imitators,

35

or possibly that banks tend to concentrate in their own products at later stages of the innovation sequence. Here we have seen evidence supporting the former, although not against the latter. In any case, it is quite apparent that the innovative process evolves to a situation where the top tier banks have all learnt enough about the type of products in a group, and most of them underwrite deals with their own dierentiated innovations. At this stage, a comparison can be made with what we know of corporate underwriting in standard equity. There, with little room for dierentiation in the security design, banks are dierentiated by the coverage of their analysts or their reputation as underwriters of high quality stocks. Also, common stock underwriting exhibits a great deal of cooperation between banks that need each others clienteles to sell their issues successfully. Further, underwriting deals are priced almost symmetrically. Thus, it is clear that the determinants of market shares in both lines of business are quite dierent. But even if this study of corporate derivatives has shown a more competitive side of investment banks, it is most likely the case that the two lines of business are related. To what extent the banks success in one of them may aect the success in the other remains an open question. A fertile ground for future research may consist of a study of the investment bank as a multiproduct rm, where innovation in corporate products allows the bank to increase its prot margins by reaching larger deals, but cooperation with other banks in more conventional securities markets is crucial to the bank to develop the necessary reputation. The ndings here also motivate future research on how the market structure aects the speed of nancial innovation. The benets of innovation are decreasing, as imitators catch up faster along the sequence. Yet more innovation feeds back into the demand for future generations. Are the incentives to innovate increasing or decreasing along the sequence? Future research may look into the determinants of the speed and duration of innovation across dierent product groups, including certainly other classes of securities, not only equity-linked derivatives.

Summary and Discussion

This paper has provided new evidence of the sources of rst-mover advantages in innovations in nance. The existing empirical literature of nancial innovation identied the following stylized fact: that investment 36

banks are able to prot from innovation despite being imitated almost immediately. Whatever advantage they had over competitors, the clue to the protability of unpatentable innovation in nance was that innovators were able to underwrite the largest market shares of corporate initial oerings. This paper has tried to provide an answer to the question of what is the source of the advantage. For this purpose I used data of all the New Issues using Equity-Linked and derivative corporate securities. This paper has tested empirically the hypothesis that rms have stronger preferences for underwriters that are innovators, not imitators. The theoretical motivation for this conjecture was the following: rms that need to raise capital have to use a security which is engineered by investment banks that act as underwriters. If the underwriter is the innovator of the security, this signals he is better informed about the choices that will be best for the rm. On average, the value to the rm from doing the issue with the innovator will be larger. To nd an appropriate method to test this hypothesis I started by analyzing preliminary evidence that suggested that innovations in corporate products such as equity-linked securities are frequently improvements or generations of previous designs, so that families of securities could be identied. I also noted that banks oered dierentiated underwriting services. Thus, I used the discrete choice theory of product dierentiation as the framework to model the decisions of rms to choose security structures and underwriters. The evidence also suggested that innovators had advantages that presumably dissipated over time. Thus, I decided not only to study the overall advantage of innovators, but its dynamics. For that purpose I specied the value to a rm for choosing a particular security and a particular underwriter whose parameters were estimable. I claimed that the advantage that the innovator had over its competitors in the market to underwrite new issues can be summarized in an index that included his identity, the time elapsed after the innovator was imitated, and the generation of a security. Moreover, this index appeared directly in the value function of a rm because banks make dierent engineering choices contingent on their private information. Using data of all the new issues of corporate securities from the Securities Data Company Database I estimated the parameters of the dynamics of the innovators advantage for multinomial logit and nested logit demand models. I also used nancial data from COMPUSTAT about the rms that issued the securities in the sample to enrich the specication. A result consistent to all the specications was that preferences for

37

innovators are, on average, stronger than for imitators. Interestingly, these preferences were initially stronger the later generation of an innovation, possibly reecting the fact that late generations get more complex and are therefore harder to understand to imitators. The preference for an innovator over an imitator diminishes in time, possibly as a result of imitators catching up with innovators. Further, the speed of the reduction in the preference for innovators over imitators was larger for later generations. I interpreted this as the fact that late generations appear naturally when more information has been aggregated about the family of securities they belong to, making learning about the innovators private information easier. This evidence here has also shown a dierent face of the investment banking industry. As opposed to common stock underwriting, there is a large potential for product dierentiation, and little incentives to cooperate with other banks so as to keep the demand advantages. Imitation, rather, is crucial for banks to catch up and achieve a high level of competitiveness. However, it is possible too that the banks success in more traditional lines of business may aect the success in markets with innovation. Asking how this is so may award research in the future. Also, as the benets of innovation are decreasing but as innovation feeds back into the demand for future generations through reputation, it is worth asking if the incentives to innovate increase or decrease along the sequences of innovations. The scope of the paper has been limited by the availability of data. Cost data was unavailable for most of the observations, making it unworthy to estimate the model jointly with a pricing equation. This would have also allowed to test if innovators and imitators have dierent marginal costs for underwriting oers, another potential source of rst-mover advantages. This paper has also taken innovation as exogenous. The set of choices available to rms was taken as given at each time. Certainly, one interesting way to continue this line of research would be to identify the preferences of rms for new securities at each time they make their choices. If the choices of the rm were to choose a security of a set of already existing securities or to rather choose to be the rst issuer of a new security, then the data in each deal could reveal what determines when an innovation is to be introduced.

38

Figure 1
This Figure shows the plots of the total principal underwritten using dierent types of perpetual Preferred Stock whose dividend payments are used by the issuer as deductions to taxable earnings. Underwritten totals are aggregated on quarterly bases and expressed in millions of current US dollars. MIPS are cumulative guaranteed monthly income preferred. QUIPS are quarterly version of MIPS, and QUICS are debt products, convertible to preferred at the option of the investor. TOPRS pay cumulative monthly cash distributions, and are issued by trusts, not limited liability corporations. The rst generation product of this family were MIPS. Dierentiated versions of MIPS appear later as next generations and their use as corporate nance instruments also peaks later.

Figure 2
This Figure shows the plots of the total principal underwritten using dierent types of mandatorily convertible Preferred Stock. Underwritten totals are aggregated on quarterly bases and expressed in millions of current US dollars. PERCs are preferred stock convertible in 3 years, paying a premium dividend, and callable anytime before conversion. DECS are similar, but the issuer gives up the call option before conversion. STRYPES are a later generation of this family, and are convertible to cash also. PEPS are 4 year convertibles that are also puttable. Dierentiated versions of PERCS appear later as next generations and their usage also peaks later.

Figure 3
This Figure shows the plots of the total principal underwritten by dierent Investment banks using 3-year mandatorily convertible preferred stock (PERCS), accumulated on a daily basis. Morgan Stanley was the innovator of PERCS, and Dean-Witter (prior to the merger) and Merrill Lynch its imitators. PERCS are a rst generation product, i.e., the rst product in the family of derivatives of convertible preferred stock.

Figure 4
This Figure shows the plots of the total principal underwritten by dierent Investment banks using noncallable mandatorily convertible preferred stock (DECS), accumulated on a daily basis. Salomon Brothers 39

was the innovator and it was imitated by Lehman. DECS are a third generation product, i.e., the third product to appear in the family of derivatives of convertible preferred stock.

Figure 5
This Figure shows the plots of the total principal underwritten by dierent Investment banks using TrustOriginated, 3-year mandatorily convertible preferred stock, accumulated on a daily basis. This security is a 14th generation product, i.e., the 14th product to appear in the family of derivatives of convertible preferred stock. The innovator is Robertson Securities, and it is imitated by Salomon, Morgan Stanley - Dean Witter among others.

Figure 6
This Figure shows the plots of the total principal underwritten by dierent Investment banks using bonds with face value tied to appreciation of hard currencies (Principal Exchange Rate-Linked Securities, PERLS), accumulated on a daily basis. Morgan Stanley was the innovator and it was imitated by Goldman, Sachs. PERLS are a rst generation product, i.e., the rst product to appear in the family of debt whose face value is tied to a publicly observable index.

Figure 7
This Figure shows the plots of the total principal underwritten by dierent Investment banks using bonds with face value tied to appreciation of S&P 500 or Amex Oil Index (SMARTS), accumulated on a daily basis. Merrill Lynch was the innovator and it was imitated by Goldman, Sachs and Lehman. SMARTS are a 4th generation product, i.e., the fourth product to appear in the family of debt whose face value is tied to a publicly observable index.

Figure 8
This gure illustrates the decision tree faced by an issuer of a corporate security, indexed by i . It chooses one variety, j 2 f 1; 2 ; :::; J g , which is valued as u ij : A variety is determined by a combination of a security type, g 2 f1 ; :::; Gg and a banker within the type. 40

Figure 9
This gure plots the estimated dynamics of the demand advantage that an innovator has over its imitators. The vertical axis measures the parametric shift in the demand function for an innovator-engineered security relative to an imitator-engineered security. It is expressed as the predicted ratio of the innovators market share to the imitators market share:
sinnovator : simitator

The horizontal axis measures the time elapsed after the rst

imitative deal any given security, in years. Each plotted line shows the estimated advantage dynamics for a given generation. For a given time horizon, the predicted advantage to the innovator decreases monotonically in the generation number. The advantage estimates are derived from the parameter estimates of the nested logit demand, aggregating the data for through 8 time periods.

Figure 10
This gure plots the estimated dynamics of the demand advantage that an innovator has over its imitators. The vertical axis measures the parametric shift in the demand function for an innovator-engineered security relative to an imitator-engineered security. It is expressed as the predicted ratio of the innovators market share to the imitators market share:
sinnovator simitator :

The horizontal axis measures the time elapsed after the rst

imitative deal any given security, in years. Each plotted line shows the estimated advantage dynamics for a given generation. For a given time horizon, the predicted advantage to the innovator decreases monotonically in the generation number. The advantage estimates are derived from the parameter estimates of the nested logit demand, aggregating the data for through 11 time periods.

Figure 11
This gure plots the estimated dynamics of the demand advantage that an innovator has over its imitators. The vertical axis measures the parametric shift in the demand function for an innovator-engineered security relative to an imitator-engineered security. It is expressed as the predicted ratio of the innovators market share to the imitators market share:
sinnovator : simitator

The horizontal axis measures the time elapsed after the rst

imitative deal any given security, in years. Each plotted line shows the estimated advantage dynamics for a given generation. For a given time horizon, the predicted advantage to the innovator decreases monotonically

41

in the generation number. The advantage estimates are derived from the parameter estimates of the nested logit demand, aggregating the data for through 12 time periods.

Figure 12
This gure plots the estimated dynamics of the demand advantage that an innovator has over its imitators. The vertical axis measures the parametric shift in the demand function for an innovator-engineered security relative to an imitator-engineered security. It is expressed as the predicted ratio of the innovators market share to the imitators market share:
sinnovator : simitator

The horizontal axis measures the time elapsed after the rst

imitative deal any given security, in years. Each plotted line shows the estimated advantage dynamics for a given generation. For a given time horizon, the predicted advantage to the innovator decreases monotonically in the generation number. The advantage estimates are derived from the parameter estimates of the nested logit demand, aggregating the data for through 16 time periods.

Figure 13
This gure plots the estimated dynamics of the demand advantage that an innovator has over its imitators. The vertical axis measures the parametric shift in the demand function for an innovator-engineered security relative to an imitator-engineered security. It is expressed as the predicted ratio of the innovators market share to the imitators market share:
sinnovator simitator :

The horizontal axis measures the time elapsed after the rst

imitative deal any given security, in years. Each plotted line shows the estimated advantage dynamics for a given generation. For a given time horizon, the predicted advantage to the innovator decreases monotonically in the generation number. The advantage estimates are derived from the parameter estimates of the nested logit demand with issuers heterogeneity, aggregating the data for through 8 time periods.

42

Appendix 1
Proof of Lemma 1. The value of choosing variety j is u ( j ; :) + " ij : Thus, given the properties of u(:); if j maximizes u ijt then it must be that

8m

@u (:) j = u m ( j ; : ) = 0; j @m j

(18)

where m j corresponds to each entry of the j vector. To nd the prot maximizing choices, let us rst solve for the demand function for some variety j , as given in (3). Note rst that the probability that an arbitrary issuer chooses variety j over any other variety k is

Pr( uij

uik ) = Pr("ik u ( j ) u ( k ) + "ij ) G(u ( j ) u( k ) + " ij ) :

Further, since each " ik is drawn independently from G( :); the probability that j is the chosen variety for i is Y

j 6=k G( u( j )

u ( k ) + "ij );

and the aggregate demand for this variety is just Z Y

"

qj ( j ; :) = M

k 6=j G(u ( j )

u( k ) + " 0)dG ("0 ):

ej would solve the set of rst-order conditions A prot maximizing choice of 8m @ j je j = 0: @ m j

43

Note that any of these rst-order conditions have the form:

@ j @qj ( j ; :) @qj ( j ; :) c0 (qj ( j ; :)) ; m = pj m @ j @ j @m j where Z

@qj ( j ; :) =M @ m j

"

um ( j )f

X G0 (u ( j ) u ( k ) + "0 ) Y g G (u ( j ) u( k ) + "0 )dG ("0 ): G( u( j ) u ( k ) + "0 )

k 6=j k 6=j @q j ( j ;:) @ m j

Thus, since

@ j @m j

j ( j ;:) = f p j c0 (qj ( j ; :))g @ q@ and p c0 > 0 , it must be that m j

= 0 if each m j is a

maximizer. Moreover, Z

@qj ( j ; :) = M um ( j ) @ m j

"

um ( j )f

X G0 (u ( j ) u( k ) + " 0) Y g G (u( j ) u ( k ) + "0 )dG( "0 ) G(u ( j ) u ( k ) + "0)

k 6=j k 6=j @q j ( j ;: ) @m j

where G0 (:) > 0 ; G (:) > 0 so the integral above is strictly positive. u m ( j ) = 0 8 m and that is only when j = j:

= 0 is only satised when

44

Proof of Proposition 1. Let j and k; l be varieties of the same security g , and q an arbitrary variety of another arbitrary security. By Lemma 2

v j ( y p; 1) v j ( y p; 1)

vk (y p; 0); vl (y p; 0):

The aggregate demand for the innovators variety, given than varieties of the same security are priced symmetrically is Z

qj (p; :) =

G (vg + "0 ) Y

"

l6= k; l2Jg q 6=j;q 2J = g G(v (y

G (v (y p; 1) v (y p l ; 0) + "0 )

p; 1) v (y pq ; b ) + "0 )dG ("0 ):

For the imitators varieties, the demand is Z

qk ( p; :) =

"

G( vg + "0 )

l 6=k;l2J g

G (v (y p; 0; :) v ( y pl ; 0 ; :) + "0 )

q 6=j; q 2J = g G (v (y

p; 0) v (y p q ; b) + "0 )dG(" 0) :

Clearly, since G (:) is a strictly increasing function, by Lemma 2

G(v g + "0 ) G(v (y p; 1) v (y p l ; 0) + "0 )

G( vg + "0 );

G( v (y p; 0) v (y p l ; 0) + "0 ); and G(v (y p; 1) v (y pq ; b ) + "0 ) G( v (y p; 0) v (y p q; b ) + " 0) :

45

and thus

qj (p ; :) qk (p ; :) :

Note that if v g > 0 ; then qj (p ; :) >qk (p ; :).

46

Proof of Lemma 3. The proof follows directly from the fact that G (:) is a strictly increasing function. If vg is decreasing with time, then so is the dierence between G(v g + "0 ) and G( vg + " 0).. The aggregate demand qj (p ; :) is also strictly increasing in G(:); so qj (p; :)qk (p; :).is decreasing in vg ; and thus decreasing in time.

47

Appendix 2
List and brief description of the Equity-Linked and Derivative securities in the SDC Database from 1985 to 2001
Name Description

1. Debt Products RISRS Rising Interest Subordinated Redeemable Securities. Debt notes subordinated to existing and future debt notes.

2. Convertible Debt (Zero Coupon) LYONS Liquid Yield Option Notes. Zero-Coupon b onds, convertible to common, stock, temporarily non-callable and non-puttable. 3. Convertible Debt (Dividend Paying) SIRENS Set-Up Income Redeemable Equity Notes. Low dividend paying, subordinated debt, callable, convertible to common stock. ICONS Investment Company Convertible Notes. Notes issued by closed-end mutual funds that are puttable, conversion price adjusted yearly.

4. Convertible Preferred PERCS Preferred Equity Redemption Cumulative Stock. Mandatorily convertible preferred, after 3 years, callable anytime. YES Shares Yield Enhanced Stock. High dividend paying preferred, mandatorily convertible in 3 years, callable anytime. (continues)

48

Name

Description

4. Convertible Preferred (contd.) DECS Dividend Enhanced Convertible Stock. Non-callable, mandatorily convertible preferred stock. X-Caps Exchangeable Capital Securities. Fixed dividend paying, perp etual, callable, convertible to subordinated preferred stock. ACES Automatically Convertible Equity Securities. Puttable, mandatorily convertible to common stock in 4 years. PRIDES Preferred Redeemable Increased Dividend Equity Securities. Callable, non-puttable, 4 years conversion, premium dividend paying preferred stock. PEPS Participating Equity Preferred. High dividend yield, 4 year conversion, callable and puttable preferred stock. SAILS Stock Appreciation Income Linked Securities. 3 years mandatory conversion, converted at premium if common stock appreciates. AutoComEx Automatic Common Exchange Securities. Treasury notes, mandatorily convertible to issuers common in 3 years. Callable. PEPS (2) Premium Equity Participation Securities. Preferred Stock mandatorily convertible in 3 years, conversion ratio determined ex-post. STRYPES Structured Yield Product Exchangeable. Preferred stock convertible to common or cash. Initial 20% appreciation given up in exchange of higher yield. MARCS Mandatorily Adjustable Redeemable Preferred. Callable, convertible in 4 years preferred stock. Depreciation of convertible common bounded at 10%. MEDS Mandatorily Exchangeable Debt Securities. 4-year mandatorily convertible preferred at a premium. (continues)

49

Name

Description

4. Convertible Preferred (contd.) Trust Convertible Preferred Mandatorily convertible preferred, issued by a business trust, not a Limited Liability Corp oration. TRACES Trust Automatic Common Exchange Securities. Trust-originated preferred stock convertible to common stock. Conversion after 3 years.

5. Short-Term Tax-Advantage FRAPS Floating/Adjustable Rate Preferred Stock. Preferred Stock that pays a xed for 5 years. Adjustable rate after 5th year, if not called.6.

6. Perpetual Tax-Advantage MIPS Cumulative Guaranteed Monthly Income Preferred. Perpetual preferred stock. Dividend payments used as tax deductible by the issuer. EPICS Exchangeable Preferred Income Cumulative Shares. Perpetual preferred stock with the issuers option to convert to common anytime. MIDS Monthly Income Debt Securities. Perp etual preferred stock with the issuers option to defer monthly dividend payments. TOPRS Trust-Originated Preferred. 49 year preferred stock, issued by business trust. Monthly cumulative xed cash distributions. QUIDS Quarterly Income Debt Securities. 50-year preferred with b ounded deferral horizon and initially non-callable. QUIPS Quarterly Income Preferred Securities. 49-year preferred with the issuers option to shorten maturity. (continues)

50

Name

Description

6. Perpetual, Tax-Advantage (contd.) QUICS Quarterly Income Capital Securities. Tax-deductible, 30-year bonds with interest deferral option. RES-CAPS Reset Capital Securities. Perpetual preferred stock, with an initial call protection. COPRS Canadian Originated Preferred Securities. Non-convertible, tax-deductible dividend-paying preferred, issued by Canadian LLCs.

7. Convertible, Tax-Advantage Convertible MIPS Optionally convertible p erpetual preferred stock that is tax deductible. Initially protected from a call, with optional interest deferral. TECONS Term Convertible Shares. Mandatorily convertible, tax deductible, and callable only if price of common stock is high enough. Convertible TOPRS QDCS Same as TOPRS (above), but mandatorily convertible to common stock. Quarterly Debt Capital Securities. 20-year or less securities, convertible to preferred stock at maturity. Tax deductible coupon, callable. EPPICS Equity Providing Preferred Income Convertible. 40 year convertible bonds, convertible to cash, common or preferred stock. TRUPS Trust Preferred Stock. Callable notes, tax deductible, that investor has to buy jointly with preferred stock from a subsidiary of issuer. Convertible QUIPS Same as QUIPS (above), convertible to common, at dierent maturities. (continues)

51

Name

Description

8. Index-Tied Principal PERLS Principal Exchange Rate-Linked Securities. Dollar denominated bonds with redemption value tied to a hard currency . SIRS Stock Index Return Securities. Bonds with face value adjusted upwards if S&P MidCap 400 index appreciates enough. MITTS Market Index Target-Term Securities. Zero coupon bond, principal tied to the S&P 500 Index. SMARTS Stock Market Annual Reset Term Notes. Variable capped coupon paying bond, principal can be tied to S&P or Amex Oil indexes. EPS Equity Participation Securities. Zero coupon bonds. Appreciation of principal pegged to S&P 500, capp ed and oored. CPNs Currency Protected Notes. Principal of bonds tied inversely to Canadian sixmonth bankers acceptance rate. SUNS Stock Upside Note Securities. Principal tied to a basket of telecom stocks. The depreciation is oored. CUBS Customized Upside Basket Securities. Zero coup on, downside protected bond with principal tied to a basket of stocks across industries.

9. Stock Tied-Principal ELKs Equity-Linked Securities. Annual interest paying notes. Principal tied to the performance of the stock of a chosen rm. YEELDS Yield-Enhanced Equity-Linked Debt Securities. 3-year notes. Principal tied to the stock of a rm in the information technology sector. Appreciation capped. (continues)

52

Name

Description

9. Stock-Tied Principal (contd). CHIPS Common-Linked Higher Income Participation Securities. High-yield annual coupon debt with appreciation tied to stock options, oored and capped. PERQS Performance Equity-Linked Redemption Quarterly Pay Securities. Quarterly coupon paying notes. Principal tied to a given stock at investors option.

10. Privatization Exchangeable Debt PENs Privatization Exchangeable Notes. Foreign currency and dollar denominated b onds exchangeable for a privatized rm shares.

53

Appendix 3
Summary Statistics
Variable Mean Std. Dev. Median Minimum Maximum

1. Oer Principal (US$ millions) Underwriting spread (%) Private placement (Yes = 1) Days before rst imitative deal

306:160 2:442 0:138 484:440

446 :515 1 :135 0 :345 642 :875

175 :50 3 :00 0 214

10 :00 0:085 0 7

4 ; 050:00 12:60 1 2408

2. Issuer information Total Capitalization (US$ millions) Common stock (% of capitalization) Preferred stock (% of capitalization) Long-term debt (% of capitalization) Short-term debt (% of capitalization)

15 ; 668:700 43:436 7:574 43:522 15:991

44; 750 :980 22 :513 11 :501 19 :546 16 :551

4; 115 :70 44 :97 5 :29 42 :98 10 :14

1 :20 0 :27 0 :01 0 0 :01

742 ; 536:60 100 99:31 100 52:99

3. Banks market share (share of total deals, by period) Varietys share of all equity underwriting Varietys share of the product market Standard Equity market share

0 :0008 0 :6857 0 :9535

0:0019 0:3787 0:0149

0 0 :9621 0 :9454

0 0 :0256 0 :9395

0:0148 1 0:9959

(continues)

54

Variable

Mean

Std. Dev.

Median

Minimum

Maximum

4. Other variety indicators (by period) Innovations accumulated by bank in category Innovations accumulated by bank (total) Product innovator dummy (Yes = 1) Group innovator dummy (Yes = 1) Years or protection from call Yield to maturity (%) Yield advantage (%)

1 :172 5 :023 0 :585 0 :328 3 :968 6 :809 2 :151

1 :293 5 :198 0 :493 0 :470 2 :645 5 :827 3 :424

1 3 1 0 3 :963 7 :292 0

0 0 0 0 0 0 0

4 14 1 1 30 73 :094 31

5. Instruments (average over other varieties or other banks in same product market). Innovations accumulated by bank in category Innovations accumulated by bank (total) Years or protection from call Yield to maturity (%) Yield advantage (%)

0 :503 2 :566 1 :395 2 :772 0 :618

0 :743 3 :435 3 :477 4 :903 2 :0888

0 1 0 0 0

0 0 0 0 0

4 14 19 :583 24 :133 11 :998

Data are for 662 issues of equity-linked and derivative securities b etween April 1985 and March 2001. Market shares and variety specic data statistics are computed after the aggregation required by the estimation period.

55

References
Allen, F., and D. Gale (1994): Financial Innovation and Risk Sharing. MIT Press, Cambridge, MA. Anderson, S., A. DePalma, and J.-F. Thisse (1989): Demand for Dierentiated Products, Discrete Choice Models, and the Characteristics Approach, Review of Economics Studies, 56(1), 2135. Berry, S. (1994): Estimating Discrete-Choice Models of Product Dierentiation, RAND Journal of Economics, 25(2), 242262. Berry, S., J. Levinsohn, and A. Pakes (1995): Automobile Prices in Market Equilibrium, Econometrica, 63(4), 841890. Bhattacharyya, S., and V. Nanda (2000): Client Discretion, Switching Costs, and Financial Innovation, The Review of Financial Studies, 13(4), 11011127. Black, D., and W. Silber (1986): Pionnering Products: Some Empirical Evidence from Futures Markets, Salomon Brothers Center for the Study of Financial Institutions Working Paper 395, New York University. Caplin, A., and B. Nalebuff (1991): Aggregation and Imperfect Competition: On the Existence of Equilibrium, Econometrica, 59(1), 2559. Carter, R., F. Dark, and A. Singh (1998): Underwriter Reputation, Initial Returns, and the Long-Run Underperformance of IPO Stocks, Journal of Finance, 53(1), 285311. Carter, R., and S. Manaster (1990): Initial Public Oerings and Underwriter Reputation, Journal of Finance, 45(4), 10451067. Finnerty, J. (1992): An Overview of Corporate Securities Innovation, Journal of Applied Corporate Finance, 4(1), 2339. Herrera, H., and E. Schroth (2000): Protable Innovation Without Patent Protection: The Case of Credit Derivatives, Berkeley Center for Entrepreneurial Studies Working Paper 01-02, New York University. 56

Kanemasu, H., R. Litzenberger, and J. Rolfo (1986): Financial Innovation in an Incomplete Market: An Empirical Study of Stripped Government Securities, Rodney White Center Working Paper 26-86, University of Pennsylvania. Loughram, T., and J. Ritter (2004): Why Has IPO Underpricing Increased Over Time, Mimeo, University of Florida. McFadden, D. (1981): Econometric Models of Probabilistic Choice, in Structural Analysis of Discrete Data with Econometric Applications, ed. by C. Manski, and D. McFadden, pp. 198272. MIT Press. Miller, M. (1986): Financial Innovation: The Last Twenty Years and the Next, Journal of Financial and Quantitative Analysis, 21, 459471. Nanda, V., and Y. Yun (1995): Sharing the Limelight: On Why Investment Banks Co-Manage Initial Oerings of Innovative Securities with Competitors, Mimeo, University of Michigan Business School. Naslund, B. (1986): Financial Innovations: A Comparison with Research and Development in Physical Products, Salomon Brothers Center for the Study of Financial Institutions Working Paper 531, New York University. Nevo, A. (2000): A Practitioners Guide to Estimation of Random-Coecients Logit Models of Demand, Journal of Economics and Management Strategy , 9(4), 513548. Pratt, T. (1991): How Percs Became the Years Hottest Product, Investment Dealers Digest, 57(48), 20. (1995): SEC Issues No-Action Letters on Streamlined Trust MIPS, Investment Dealers Digest, 13(61). Ross, S. (1989): Institutional Markets, Financial Marketing, and Financial Innovation, The Journal of Finance, 44(3), 541556. Tufano, P. (1989): First-Mover Advantages in Financial Innovation, Journal of Financial Economics, 3(2), 350370.

57

(1995): Securities Innovations: A Historical and Functional Perspective, Journal of Applied Corporate Finance, 7(4), 90 104. VanHorne, J. (1985): Of Financial Innovations and Excesses, The Journal of Finance, 40(3), 621631.

58

Footnotes
1. For a list of innovations in corporate securities until 1991, and a description of some of them, see Finnerty (1992). All innovations in equity-linked and derivative corporate securities until March 2001 are listed in this paper. 2. The underwriting spread is the fee charged by an investment bank or a syndicate, equal to the dierence between the gross sales to investors and the net proceeds received by the issuer. 3. Tufanos study, for example, compares spreads only between banks issuing the same security. In that sense, it is not a demand estimation for the underwriting services of banks across dierent securities. 4. Tufano (1995) and Finnerty (1992) also describe the reasons for the appearance of the most important innovations between 1830 and 1930 and since the 1970s, respectively. 5. According to anecdotal evidence gathered by Naslund (1986), though, rms usually turn to the services of expert issuers of innovative products, i.e., innovators, even if they used another bank for other services. This evidence comes from the testimonies of twenty nancial product developers in New York. 6. Black and Silber (1986) also study rst-mover advantages in nance but they focus in futures exchanges as the innovators, not in investment banks. They claim that futures exchanges that develop new contracts have the advantage that they provide liquidity for investors earlier than the competing exchanges so they are able to attract agents that have to choose where to trade. 7. Kanemasu, Litzenberger and Rolfo (1986) observe the same pattern for the case of stripped treasury securities. 8. See Pratt (1995) for a detailed comparison of MIPS and TOPRS. 9. For every product I have compiled articles in Investment Dealers Digest, American Banker, Dow Jones Newswires and others found using the ABI Search Engine. For every one I was able to nd a description of the product, and a reference to an older product which was similar to it. I am especially indebted to Tom Pratt, who writes a descriptive article of almost every corporate security invented. 59

10. We can nd some attempts in the literature to endogenize the advantage of an innovator over its imitators. Herrera and Schroth (2000) provide a model that explains why innovators of derivatives acquire superior information over competitors. In a dierent perspective, Black and Silber (1986) derive an advantage to pioneering exchanges that establish liquidity early and Allen and Gale (1994, Chapter 4) show how rst-movers can preempt imitation choosing their initial capacity as a Stackelberg leader. 11. Clearly, the set of new possible engineering choices is tied to the discovery of rmss preferences for some previously non-existent feature of available security. This was certainly the case of the PERCs, where its innovator, Morgan Stanley, gured out that in the uncertain environment for investment in late 1991, an issue of a PERC would allow rms to attract investors with stable high yields, pricing the oer better than common stock while capping the securitys appreciation potential (Pratt, 1991). 12. Some securities can also be grouped into categories of similar instruments, i.e., what I referred to above as families. For example, SIRENS or ICONS are dividend paying convertible debt instruments, while ELKS or YEELDS are zero-coupon instruments with principal payment tied to the appreciation of a given stock portfolio. I do not deal with this prior level of classication in the theoretical model but we do account for proximity of securities between dierent groups in the econometric model. 13. The reader may have noticed already that this approach is equivalent to the derivation of an indirect utility function. 14. Sucient conditions for strategic complementarity would be that
@ 2qj @p j @p k

> 0 and

@q j @ pk

> 0: The latter

condition is obvious just by inspecting the aggregate demand (see appendix). The former will be met if v (:) is not too concave. 15. The sucient conditions for this would be that
@q j @v

> 0 and

@ 2qj : @q j @ v

Again, the former holds and it

is veried just by inspecting the aggregate demand function. The latter condition is met only under certain assumptions about G( :) and v (:) : 16. By imposing the quasi-linearity of income, wealth eects are ruled out ( would be the (constant) marginal utility of income). This assumption is not only tractable and convenient for the estimation 60

but quite reasonable for some types of products. I believe this assumption is justied for this particular data set: the amount paid in underwriting fees is small relative to the value of the outstanding equity of issuers (the 95th percentile of the ratio of fees to equity is 0.02) or to their yearly earnings (the median proportion of fees of the yearly earnings is 0.06; the third quartile is 0.21). As an example, Nevo (2000) argues that it is reasonable to assume quasi-linearity for ready-to-eat cereal because their price represents a small share of household expenditures. By the same token, it is not reasonable to make this assumption for the demand of cars. In fact, Berry, Levinsohn and Pakes (1995) use a Cobb-Douglas utility function, i.e., the additive term for net income is log(y it p jt ): 17. See Berry (1994) for a proof. 18. Enriching the random indirect utility will allow more reasonable substitution patterns. This can be veried by inspecting the price elasticities of the model. For two varieties, j and l, the price elasticities
@sj pl @p lsj

would be given by: 8 > > 1 > p j f 1 > s j [1 + > > <

jl =

Cross-price elasticities for two dierent varieties within the same group will be dierent than cross-price elasticities for varieties across dierent groups, even if they have the same prices and market shares (provided that > 0). Still, the cross-price elasticity of one variety with varieties of other groups will be identical, even if one variety belongs to a group that is closer to the other in terms of the uses of the security. 19. In some periods, market shares of existing varieties are zero. The computation of logit and nested logit models requires taking the logarithm if these shares. To avoid the indeterminacy problem I use a

P pl s l f1 + 1 D 1 f 2G D1 for j 6= l; and j; l 2 Jg ; f g > > > > > > : pl s l otherwise.

1 1 Dg

f 2G

D1 ]g for j = l 2 Jg ; f

(19)

61

transformation sj that does not alter its distribution signicantly. I compute instead

ln s jt ln s j=g

= ln(s jt + 0:00001); = ln(s j=g t + 0:00001):

Excluding these observations would not only bias the sample selection but also imply a loss of 191 observations that actually reveal that the demand was zero. 20. For periods of zero market shares no information of the bids made by potential underwriters is observed. Since these varieties were available to rms, although none chose to use them, we assume there is a going price for them. We approximate this price with the last observed price for that variety. 21. Notable is the case of YES Shares, a mandatorily convertible Preferred Stock product introduced by Goldman Sachs 7 months after Morgan Stanley introduced its PERCs. These two products are virtually identical in design and the only dierence between them seems to be their brand name. 22. Alternatively, one can think of using well known underwriter reputation measures, such as those found in Loughram and Ritter (2004), which update the original prestige rankings by Carter and Manaster (1990) and Carter, Dark and Singh (1998) to the year 2000. These measures base reputation on the number of times that a bank appears higher in the list of syndicated underwriters in the prospectus list. The higher the bracket, the more shares that the bank underwrites. The reputation ranking for 1980 to 2000 is of little use in this study because it exhibits very little time and cross-sectional variation for most of the banks in this sample. Of 662 deals, 650 are underwritten by banks that scored either 8.75 or 9.0 between 1985 and 1991 and 8 or 9 between 92 to 2000. Thus, a measure that distinguishes these banks in terms of their placement ability is a measure related directly to the placement capacity in this product class. The covariance matrix estimator is robust to possible correlation of errors within the same security. While errors of varieties of the same bank may also be correlated, this may not be signicant if they are in dierent product groups. Also part of this eect should be picked up by the bank-specic regressors.

62

Estimating the same model using an estimator of the covariance matrix that considers within-bank correlation rather than within security correlation the inference on the innovators advantage index or the banks characteristics does not change for all aggregations. The estimate for becomes insignicant when t = 12 and the estimate for becomes insignicant when t = 16: 23. I tried using also the cumulative number of underwriting deals rather than the principal, but its estimated coecient was never statistically signicant. 24. Merrill Lynch (the innovator) and Salomon Smith Barney comanaged the 73rd TOPRs issue, and Goldman Sachs (innovator) and Morgan Stanley comanaged the 16th QIDS issue. In both cases, I granted each comanager with one full deal for the computation of its market share. 25. While the dataset used here overlaps little with Nanda and Yuns (1995), the results here can very well be consistent with their hypothesis. They argue that the incentives to co-manage the rst deals are smaller if innovators expect a large market for their innovation and a more dominant position in the market. This dataset does exhibit virtually no co-management, deal sizes that are signicantly larger than common stock underwriting deals and dominant positions by innovators. The fact that 14 of the 17 co-managed deals happen in 4th or later generation securities could be also indicative that co-management is more likely when in late generations, which is when the evidence shows that the dominant position of innovators is smaller. 26. The grouping of generations captures the idea that some generations may actually belong to the same cohort of designs. I tried also grouping them in ve categories: category 1 for rst generations, 2 for second and third, 3 for fourth to sixth, 4 for 7th and 8th, and 5 else. The association is still negative (i.e., later generations are imitated less frequently) but weaker (p-value is 0.183).

63

Table 1: Total Principal Underwritten in the New Issues Markets using Standard Equity and Equity-Linked Products between April 1985 and March 2001
Time Period Total Underwritten Principal Standard Equity-Linked and Equity Derivative Products Standard Equity Average Deal Size Equity-Linked and Derivative Products

1 2 3 4 5 6 7 8 9 10 11 12

64 ; 477 61 ; 034 35 ; 801 45 ; 123 61 ; 216 99 ; 507 129 ; 073 90 ; 397 139 ; 518 169 ; 569 143 ; 016 230 ; 734

3 ; 915 590 4 ; 120 9 ; 314 19 ; 517 10 ; 750 22 ; 106 22 ; 414 32 ; 357 35 ; 463 19 ; 616 20 ; 680

38 36 48 46 57 57 56 56 61 83 112 381

489 84 589 388 887 336 287 270 225 269 265 440

Figures shown are expressed in millions of current US$. The source of the data is the Securities Data Company Database of New Issues. Each one of the twelve periods is an equal sized interval b etween April 1995 and March 2001.

64

Table 2: Classication of Equity Linked and Derivative Corporate Securites

Category 1. Debt Products. 2. Convertible Debt (Zero Coupon). 3. Convertible Debt (Dividend Paying) 4. Convertible Preferreds Securities RISRS. LYONS. SIRENS, ICONS. PERCS, YES Shares, DECS, X-Caps, ACES, PRIDES, PEPS, SAILS, AutoComEx, PEPS (2), STRYPES, MARCS, MEDS, Trust Convertible Preferreds, TRACES. FRAPS. MIPS, EPICS, MIDS, TOPRS, QUIDS, QUIPS, QUICS, RES-CAPS, COPRS. Convertible MIPS, TECONS, Convertible TOPRS, QDCS, EPPICS, TRUPS, Convertible QUIPS. PERLS, SIRS, MITTS, SMARTS, EPS, CPNs, SUNS, CUBS. ELKs, YEELDS, CHIPS, PERQS. PENs.

5. Short-Term Tax-Advantage 6. Perpetual Tax-Advantage

7. Convertible, Tax-Advantage

8. Index-Tied Principal 9. Stock Tied-Principal 10. Privatization Exchangeable Debt

The list shown above contains the names of all the equity-linked and derivative corporate securities issued between 1985 and 1991. The securities are classied into categories that are also referred to as groups or families. They are listed in chronological order of appearance, which is also called the generation number.

65

Table 3: Innovators Market Shares for Imitated Securities

Share of Principal (%)

Product

Innovator

Number of Rivals

Share of Deals (%)

Leader

Leader

RISRS

Kemper Securities

43

No

50

Yes (Tied)

SIRS

Paine Webber Salomon-Smith Barney Merrill Lynch

67

Yes

90

Yes

DECS Equity Participation Securities MIDS SMART Notes X-CAPs ELKS

1 1

94 94

Yes Yes

79 50

Yes Yes (Tied) No (2) Yes Yes Yes

Goldman, Sachs Merrill Lynch Merrill Lynch Salomon-Smith Barney Morgan Stanley Goldman, Sachs Morgan Stanley Merrill Lynch Merrill Lynch Lehman Brothers Goldman, Sachs Merrill Lynch Salomon Brothers Merrill Lynch Robertson

2 2 2 2

29 60 67 82

No (2) Yes Yes Yes

30 41 92 90

PERCS QIDS PERLS LYONS FRAPS RST-CAPS MIPS TOPRS TRUPS Convertible TOPRS Trust-Originated Convertible Preferreds

2 2 2 2 3 3 3 3 4 6 7

82 82 93 97 22 25 89 95 71 47 6

Yes Yes Yes Yes No (2) Yes (Tied) Yes Yes Yes Yes No (5)

88 76 97 95 23 22 96 92 72 33 5

Yes Yes Yes Yes No (2) No (2) Yes Yes Yes No (2) No (4)

Total Number of Leads (of 18 cases)

15

14

An entry of Yes in the fth or last columns (Leader) means that the innovator had the biggest market share. A No entry is followed by the innovators ranking in brackets. A tie for the lead is also shown in brackets.

66

Table 4: Ratio of the Total Principal Underwritten by Imitators relative to the Principal Underwritten by the Innovator (Average over imitated securities) Generation of Imitated Securities Average Ratio

First Generation (7 securities) All Generations (18 securities)

0 :337 0 :679

The ratio is the total principal underwritten by the innovator divided by the total principal underwritten by all imitators, for any given security. The ratio is averaged over all the imitated securities and over only those imitated securities that were a rst generation product. There are 18 securities that were imitated, 7 of which where rst generations.

67

Table 5: Within-Security Standard Deviation of Some Security Characteristics for Innovators and Imitators Characteristics of the Oering Imitators Innovator

Oer Yield to Maturity n Spread over Treasury Bills n Years of Call Protection n Years before Called at Par n Yield Advantage n Percentage Yield n

0 :779 46 25 :420 10 4 :055 31 1 :043 23 0 :840 8 0 :858 35

12 :949 418 87 :528 31 1 :772 407 2 :703 296 3 :610 167 15 :569 288

This table reports the standard deviations of some characteristics within the security. The within variation is computed for all imitated equity-linked and derivative securities, engineerd by innovators and imitators. The number of observation used for each is denoted by n.

68

Table 6: Paired Dierence of Means Test for the Underwriting Spreads of Innovators and Imitators = Underwriting Spread by Innovator - Underwriting Spread by Imitator Periods of Aggregation Observations Mean Standard Error P-value ( > 0 )

16 12 11 8

137 113 110 79

0:058 0:063 0:051 0:083

0:0537 0:0573 0:0577 0:0706

0 :139 0 :137 0 :189 0 :121

The means of the underwriting spreads for innovators and imitators at a given time period for a given security are compared using a paired dierence of means test. Four dierent frequencies are used to dene a time period: one year (t = 16); 17 months (t = 12); 18 months (t = 11); and 2 years (t = 8): The alternative hypothesis to compute the p-value is that the dierence of means is p ositive.

69

Table 7: Estimation Results with Logit Demand

The dependent variable is the dierence of the logarithms of the market shares of a bankss underwriting of a given security and of the market share of equity underwritings: ln( sj ) ln(so ): Regressors (parameter name) Base case Parameter estimates With dynamics

Underwriting Spread ()

0 :735 (0 :441)

0:783 (0 :450) 0 :842 (0 :033) 0 :361 (0 :150) 4:433 (1 :081) 0:177 (0 :054)

Total equity-linked innovations by the underwriter Total equity-linked Innovations in that category by the underwriter Innovator dummya

0:075 (0 :031) 0:359 (0 :152) 2 :010 (0 :320)

Innovator dummy interacted with the securitys generation number Innovator dummy interacted with imitation time and generation number

0:334 (0 :105)

Group dummiesb Time dummiesc

Yes Yes

Yes Yes

Instrumented regressors

Underwriting Spread

Underwriting Spread

2 statistic for OIR Testd

P-value

14 :115 0 :079

17 :1836 0:028

(continues)

70

Table 7: continued. The dependent variable is the dierence of the logarithms of the market shares of a bankss underwriting of a security and of the market share of equity underwritings: ln(sj ) ln(so ): Regressors (parameter name) (continued) Average elasticity of market share to the Underwriting Spread Median elasticity of market share to the Underwriting Spread Number of predicted inelastic demands Base case Parameter estimates With dynamics

1 :834 (0 :797) 2 :018 50

1 :955 (0:850) 2 :151 49

Number of observations R-squared F-statistic

323 0 :292 45:47

323 0 :324 11:85

The sample includes all oerings of equity-linked and derivative securities b etween 1985 and 2001. The parameters of a theoretical logit demand model are obtained using an Instrumental Variables estimator. The standard errors are obtained using a Hub er-White covariance estimator, robust for heteroskedasticity and within-security correlation. They are shown below the estimate, in parenthesis. Estimates followed by level, by

are signicant to the 0.01

to the 0.05 level, and by

to the 0.1 level. Each observation consists of the market share of a banks

variety of a given security at a given time period, associated with the characteristics of the security and the banker, in the same p eriod, and their instruments. The sample time span has b een split in 12 p eriods of equal length.
a

Equals one when the underwriter is the innovator of that security. A dummy for each one of the 10 categories. Equals one when the security belongs to that category. A dummy for each one of the 12 periods. Equals one when the observation corresponds to that period. Test of Over-identied Restrictions. The null hyp othesis is that the model is correctly specied and that the instruments over-identify the variation in the endogenous regressor. The p-value is shown below the 2 statistic.

71

Table 8: IV estimates for the base Nested Logit Demand Model

The dependent variable is the dierence of the logarithms of the market shares of a bankss underwriting of a given security and of the market share of equity underwritings: ln( sj ) ln(so ): Regressors (parameter name) Parameter estimates

t = 16

t = 12

t = 11

t=8

Underwriting Spread ()

1 :858 (0 :919)

1:025 (0 :560) 0:658 (0 :367) 0 :088 (0 :033) 0:386 (0 :154) 2:943 (1 :322) 0 :171 (0 :056) 0 :291 (0 :106)

1 :206 (0:577) 0:618 (0:359) 0:096 (0:034) 0:422 (0:158) 2:532 (1:341) 0:123 (0:040) 0 :270 (0:119)

1:939 (1 :246) 0 :745 (0 :300) 0:034 (0 :042) 0:104 (0 :208) 2:976 (1 :835) 0:129 (0 :050) 0:476 (0 :250)

Product Market Share, in logarithms ( )

0 :634 (0 :285)

Total equity-linked innovations by the underwriter Total equity-linked Innovations in that category by the underwriter Innovator dummya

0 :085 (0 :033) 0:515 (0 :172) 2 :701 (1 :276)

Innovator dummy interacted with the securitys generation number Innovator dummy interacted with imitation time and generation numbers

0:116 (0 :038) 0 :157 (0 :088)

Group dummiesb Time dummiesc Instrumented regressors: Underwriting Spread and Product Market Shares (in logarithms)

Yes Yes

Yes Yes

Yes Yes

Yes Yes

2 statistic for OIR Testd

P-value

3:553 0:981

32 :300 0 :000

27:206 0:004

3:015 0:991
(continues)

72

Table 8: continued. The dependent variable is the dierence of the logarithms of the market shares of a bankss underwriting of a security and of the market share of equity underwritings: ln(sj ) ln(so ): Regressors (parameter name) (continued) Average elasticity of market share to the Underwriting Spread Median elasticity of market share to the Underwriting Spread Number of predicted inelastic demands Parameter estimates

t = 16

t = 12

t = 11

t=8

12:967 (5 :542) 14:398 9

8:276 (3:596) 9:100 19

7 :958 (3:440) 8 :904 10

18 :778 (8:541) 21 :497 4

Number of observations R-squared F-statistic

418 0:093 9 :04

323 0:290 12 :30

312 0 :258 10:33

225 0 :029 8:55

The sample includes all oerings of equity-linked and derivative securities b etween 1985 and 2001. The parameters of a theoretical logit demand model are obtained using an Instrumental Variables estimator. The standard errors are obtained using a Hub er-White covariance estimator, robust for heteroskedasticity and within-security correlation. They are shown below the estimate, in parenthesis. Estimates followed by level, by

are signicant to the 0.01

to the 0.05 level, and by

to the 0.1 level. Each observation consists of the market share of a banks

variety of a given security at a given time period, associated with the characteristics of the security and the banker, in the same p eriod, and their instruments. The sample time span has b een split in 16,12,11 and 8 periods.
a

Equals one when the underwriter is the innovator of that security. A dummy for each one of the 10 categories. Equals one when the security belongs to that category. A dummy for time period. Equals one when the observation corresponds to that period. Test of Over-identied Restrictions. The null hyp othesis is that the model is correctly specied and that the instruments over-identify the variation in the endogenous regressor. The p-value is shown below the 2 statistic.

73

Table 9: IV estimates for the Nested Logit Demand with Firm Heterogeneity
The dependent variable is the dierence of the logarithms of the market shares of a bankss underwriting of a given security and of the market share of equity underwritings: ln( sj ) ln(so ): Regressors (parameter name) Parameter estimates

t = 16

t = 12

t = 11

t=8

Underwriting Spread ()

0 :605 (0 :579)

1:041 (0:539) 0:664 (0:347) 1 :91 (0:500) 0:023 (0:015) 4 :679 (1:625) 0:188 (0:065) 0:467 (0:137)

0 :987 (0 :581) 0 :821 (0 :461) 1:26 (0 :743) 0 :018 (0 :014) 3:841 (1 :698) 0:120 (0 :053) 0 :466 (0 :149)

1:782 (1 :027) 0:558 (0 :383) 3 :07 (0 :836) 0:056 (0 :028) 7:022 (2 :066) 0:199 (0 :069) 1:001 (0 :258)

Product Market Share, in logarithms ( )

0 :644 (0 :396)

Underwriting Spread interacted with Market Capitalization (US$ Million) Underwriting Spread interaction with Propotion of Preferred Stock Innovator dummya

1:11 (0 :415) 0 :022 (0 :015) 4 :980 (1 :370)

Innovator dummy interacted with the securitys generation number Innovator dummy interacted with imitation time and generation number

0:111 (0 :047) 0 :359 (0 :075)

Group dummiesb Time dummiesc Instrumented regressors: Underwriting Spread and Product Market Share (in logarithms)

Yes Yes

Yes Yes

Yes Yes

Yes Yes

2 statistic for OIR Testd

P-value

2 :556 0 :995

15:660 0:154

11 :975 0 :366

2:974 0:992
(continues)

74

Table 9: continued. The dependent variable is the dierence of the logarithms of the market shares of a bankss underwriting of a security and of the market share of equity underwritings: ln(sj ) ln(so ): Regressors (parameter name) (continued) Average elasticity of market share to the Underwriting Spread Median elasticity of market share to the Underwriting Spread Number of predicted inelastic demands Parameter estimates

t = 16

t = 12

t = 11

t=8

4 :341 (1:939) 4 :666 18

7 :717 (3 :490) 8 :260 10

13 :879 (6 :296) 15 :061 7

9 :876 (4:722) 10 :751 3

Number of observations R-squared F-statistic

332 0 :354 113:19

261 0 :369 41:71

250 0 :304 55 :23

177 0 :290 10:19

The sample includes all oerings of equity-linked and derivative securities b etween 1985 and 2001. The parameters of a theoretical logit demand model are obtained using an Instrumental Variables estimator. The standard errors are obtained using a Hub er-White covariance estimator, robust for heteroskedasticity and within-security correlation. They are shown below the estimate, in parenthesis. Estimates followed by level, by

are signicant to the 0.01

to the 0.05 level, and by

to the 0.1 level. Each observation consists of the market share of a banks

variety of a given security at a given time period, associated with the characteristics of the security and the banker, in the same p eriod, and their instruments. The sample time span has b een split in 16,12,11 and 8 periods.
a

Equals one when the underwriter is the innovator of that security. A dummy for each one of the 10 categories. Equals one when the security belongs to that category. A dummy for each time period. Equals one when the observation corresponds to that period. Test of Over-identied Restrictions. The null hyp othesis is that the model is correctly specied and that the instruments over-identify the variation in the endogenous regressor. The p-value is shown below the 2 statistic.

75

Table 10: IV estimates for the Nested Logit Demand Model excluding Trademarked Securities
The dependent variable is the dierence of the logarithms of the market shares of a bankss underwriting of a given security and of the market share of equity underwritings: ln( sj ) ln(so ): Regressors (parameter name) Parameter estimates

t = 16

t = 12

t = 11

t=8

Underwriting Spread ()

1:833 (1 :030)

0 :825 (0 :512) 0 :654 (0 :362) 0 :086 (0 :033) 0 :468 (0 :151) 3:215 (1 :352) 0:144 (0 :058) 0 :308 (0 :109)

0:999 (0:513) 0 :704 (0:353) 0:092 (0:034) 0:497 (0:156) 2:532 (1:342) 0 :102 (0:042) 0 :273 (0:120)

1:282 (0 :995) 0 :667 (0 :289) 0:033 (0 :039) 0:175 (0 :201) 3 :620 (1 :661) 0 :105 (0 :050) 0 :538 (0 :233)

Product Market Share, in logarithms ( )

0 :697 (0 :323)

Total equity-linked innovations by the underwriter Total equity-linked Innovations in that category by the underwriter Innovator dummya

0 :069 (0 :034) 0:553 (0 :182) 2:648 (1 :369)

Innovator dummy interacted with the securitys generation number Innovator dummy interacted with imitation time and generation numbers

0:076 (0 :043) 0:140 (0 :100)

Group dummiesb Time dummiesc Instrumented regressors: Underwriting Spread and Product Market Shares (in logarithms)

Yes Yes

Yes Yes

Yes Yes

Yes Yes

2 statistic for OIR Testd

P-value

2:751 0:839

35 :803 0 :000

29:170 0:000

3:323 0:650
(continues)

76

Table 10: continued. The dependent variable is the dierence of the logarithms of the market shares of a bankss underwriting of a security and of the market share of equity underwritings: ln(sj ) ln(so ): Regressors (parameter name) (continued) Average elasticity of market share to the Underwriting Spread Median elasticity of market share to the Underwriting Spread Number of predicted inelastic demands Parameter estimates

t = 16

t = 12

t = 11

t=8

15:587 (6 :754) 17:258 8

5:996 (2:658) 6:796 8

8 :550 (3:759) 9 :612 8

9 :627 (4:364) 9 :627 4

Number of observations R-squared F-statistic

377 0:108 8 :79

291 0:359 12 :12

282 0 :319 10:75

204 0 :305 11:55

The sample includes all oerings of equity-linked and derivative securities b etween 1985 and 2001, except for the securities whose name was registered as a trademark in the US Patents and Trademarks Oce . The parameters of a theoretical logit demand model are obtained using an Instrumental Variables estimator. The standard errors are obtained using a Hub er-White covariance estimator, robust for heteroskedasticity and within-security correlation. They are shown below the estimate, in parenthesis. Estimates followed by level, by

are signicant to the 0.01

to the 0.05 level, and by

to the 0.1 level. Each observation consists of the market share of a banks

variety of a given security at a given time period, associated with the characteristics of the security and the banker, in the same p eriod, and their instruments. The sample time span has b een split in 16,12,11 and 8 periods.
a

Equals one when the underwriter is the innovator of that security. A dummy for each one of the 10 categories. Equals one when the security belongs to that category. A dummy for each time period. Equals one when the observation corresponds to that period. Test of Over-identied Restrictions. The null hyp othesis is that the model is correctly specied and that the instruments over-identify the variation in the endogenous regressor. The p-value is shown below the 2 statistic.

77

Table 11: IV estimates for the Nested Logit Demand Model excluding Trademarked Securities (I I)
The dependent variable is the dierence of the logarithms of the market shares of a bankss underwriting of a given security and of the market share of equity underwritings: ln( sj ) ln(so ): Regressors (parameter name) Parameter estimates

t = 16

t = 12

t = 11

t=8

Underwriting Spread ()

1 :368 (0 :602)

0 :800 (0 :492) 0 :423 (0 :245) 0 :083 (0 :018) 2:588 (1 :164) 0 :216 (0 :057) 0 :319 (0 :102)

0:788 (0:491) 0 :459 (0:229) 0:084 (0:018) 2:158 (1:152) 0 :148 (0:041) 0:309 (0:113)

1:152 (0 :994) 0:488 (0 :283) 0:080 (0 :023) 3 :481 (1 :538) 0:158 (0 :049) 0 :602 (0 :211)

Product Market Share, in logarithms ( )

0:395 (0 :239)

Total principal underwritten in the Equity Linked class by the underwriter (US$ M) Innovator dummya

0:084 (0 :016) 2 :445 (1 :062)

Innovator dummy interacted with the securitys generation number Innovator dummy interacted with imitation time and generation numbers

0:139 (0 :039) 0:203 (0 :072)

Group dummiesb Time dummiesc Instrumented regressors: Underwriting Spread and Product Market Shares (in logarithms)

Yes Yes

Yes Yes

Yes Yes

Yes Yes

2 statistic for OIR Testd

P-value

21:545 0:002

36 :249 0 :000

32:454 0:000

6:380 0:271
(continues)

78

Table 11: continued. The dependent variable is the dierence of the logarithms of the market shares of a bankss underwriting of a security and of the market share of equity underwritings: ln(sj ) ln(so ): Regressors (parameter name) (continued) Average elasticity of market share to the Underwriting Spread Median elasticity of market share to the Underwriting Spread Number of predicted inelastic demands Parameter estimates

t = 16

t = 12

t = 11

t=8

5 :820 (2 :522) 6 :446 12

3 :487 (1 :546) 3 :953 20

3:696 (1 :625) 4:156 18

5 :628 (2:551) 6 :412 6

Number of observations R-squared F-statistic

377 0 :265 13 :26

291 0 :388 17 :82

282 0:377 16 :94

204 0 :348 14:59

The sample includes all oerings of equity-linked and derivative securities b etween 1985 and 2001, except for the securities whose name was registered as a trademark in the US Patents and Trademarks Oce . The parameters of a theoretical logit demand model are obtained using an Instrumental Variables estimator. The standard errors are obtained using a Hub er-White covariance estimator, robust for heteroskedasticity and within-security correlation. They are shown below the estimate, in parenthesis. Estimates followed by level, by

are signicant to the 0.01

to the 0.05 level, and by

to the 0.1 level. Each observation consists of the market share of a banks

variety of a given security at a given time period, associated with the characteristics of the security and the banker, in the same p eriod, and their instruments. The sample time span has b een split in 16,12,11 and 8 periods.
a

Equals one when the underwriter is the innovator of that security. A dummy for each one of the 10 categories. Equals one when the security belongs to that category. A dummy for each one of the 12 periods. Equals one when the observation corresponds to that period. Test of Over-identied Restrictions. The null hyp othesis is that the model is correctly specied and that the instruments over-identify the variation in the endogenous regressor. The p-value is shown below the 2 statistic.

79

Table 12: IV estimates for the Nested Logit Demand Model: Demand for all Banks Products within a Group
The dependent variable is the dierence of the logarithms of the market shares of a bankss underwriting of all securities in a given group and of the market share of equity underwritings: ln(s j ) ln(so ): Regressors (parameter name) Parameter estimates

t = 16

t = 12

t = 11

t=8

Underwriting Spread ()

0 :825 (0 :348)

0:953 (0 :427) 0 :582 (0 :206) 0:543 (0 :215) 0 :112 (0 :025) 0 :002 (0 :012) 2:444 (1 :035) 0 :342 (0 :112)

0 :973 (0:467) 0:536 (0:187) 0:592 (0:211) 0:115 (0:028) 0:001 (0:013) 2 :203 (1:102) 0:353 (0:131)

0:711 (0 :433) 0:776 (0 :195) 0:351 (0 :225) 0:103 (0 :030) 0:005 (0 :012) 2 :495 (1 :137) 0 :466 (0 :186)

Product Market Share, in logarithms ( )

0:500 (0 :186)

Total equity-linked innovations by the underwriter Total principal underwritten in the Equity Linked class by the underwriter (US$ M) Cumulative number of deals recorded by the bank in the Equity Linked Class Innovator dummya

0 :526 (0 :205) 0:087 (0 :025) 0:018 (0 :013) 1 :946 (0 :941)

Innovator dummy interacted with the time after imitation

0:206 (0 :077)

Group dummiesb Time dummiesc Instrumented regressors: Underwriting Spread and Product Market Shares (in logarithms)

Yes Yes

Yes Yes

Yes Yes

Yes Yes

2 statistic for OIR Testd

P-value

6:755 0:914

23 :755 0 :033

15:083 0:237

6:960 0:936
(continues)

80

Table 12: continued. The dependent variable is the dierence of the logarithms of the market shares of a bankss underwriting of a security and of the market share of equity underwritings: ln(sj ) ln(so ): Regressors (parameter name) (continued) Average elasticity of market share to the Underwriting Spread Median elasticity of market share to the Underwriting Spread Number of predicted inelastic demands Parameter estimates

t = 16

t = 12

t = 11

t=8

4 :031 (1 :687) 4 :540 10

5 :554 (2 :243) 6 :221 7

5:154 (2 :099) 5:909 8

7 :609 (3:192) 8 :654 3

Number of observations R-squared F-statistic

278 0 :588 27 :80

211 0 :563 28 :48

205 0:575 23 :39

147 0 :682 24:81

The sample includes all oerings of equity-linked and derivative securities b etween 1985 and 2001. The parameters of a theoretical logit demand model are obtained using an Instrumental Variables estimator. The standard errors are obtained using a Hub er-White covariance estimator, robust for heteroskedasticity and within-security correlation. They are shown below the estimate, in parenthesis. Estimates followed by level, by

are signicant to the 0.01

to the 0.05 level, and by

to the 0.1 level. Each observation consists of the market share of a banks

variety of a given security at a given time period, associated with the characteristics of the security and the banker, in the same p eriod, and their instruments. The sample time span has b een split in 16,12,11 and 8 periods.
a

Equals one when the underwriter is the innovator of the rst security in the group. A dummy for each one of the 10 categories. Equals one when the security belongs to that category. A dummy for each time period. Equals one when the observation corresponds to that period. Test of Over-identied Restrictions. The null hyp othesis is that the model is correctly specied and that the instruments over-identify the variation in the endogenous regressor. The p-value is shown below the 2 statistic.

81

Table 13: Summary of Imitation and Innovation Facts

The sample includes all the 43 non-trademarked equity-linked and derivative securities between 1985 and 2001. Of these, 11 are a rst generation products (one per product group). There are 90 bank-security pairs (i.e., varieties) in the sample, of which 26 are rst generations. The Pearson 2 statistic is for the test with a null hypothesis that there is no statistical association between the two qualitative variables. The rows at each numeral show the data counts.

Panel A: Was the security imitated? No 1. First Generation Securities 2. Later Generations 4 21 Pearson 2 (1) = 2: 880; P-value = 0 : 090 Panel B: Was the security imitated? No 1. First Generation Securities 2. Generations 2 to 6 3. Generations 6 to 15 4 10 11 Pearson 2 (2) = 6: 124; P-value = 0 : 047 Panel C: Was the bank an imitator or innovator of previous generations in the group? No 1. Bank is not the innovator 2. Bank is the innovator 18 10 Pearson 2 (1) = 4: 064; P-value = 0 : 044 Yes 14 22 Total 32 32 Yes 7 9 2 Total 11 19 13 Yes 7 11 Total 11 32

82

Table 14: Probabilistic Regressions for the Imitation of Securities

This table shows the parameter estimates of a probabilistic regression of the probability that a security is imitated. The dependent variable is the binary indicator that takes value 1 if the security was imitated, or zero otherwise. Regressors 1 Generation Numb er Coecient estimatesa 2 3 4

0:081 (0 :022)

0:070 (0 :019) 0:569 (0 :205) 0 :527 (0 :194) 0:020 (0 :003) 0:346 (0 :068) 0 :421 (0 :193) 0 :338 (0 :155) 0:020 (0 :002) 0:302 (0 :064) 0:514 (0 :190) 0:320 (0 :182) 43 563 :85 0:281

Generation Category (1, 2 or 3)b

Total Deals Underwritten by Previous Generations

0:014 (0 :005)

0 :014 (0 :005) 0 :217 (0 :061) 0 :476 (0 :230)

Number of previous generations that were imitated

0:261 (0 :080)

Were the underwriters in this generation also of the rst generation? (1 if yes, 0 otherwise) Were the underwriters in this generation also of the second generation? (1 if yes, 0 otherwise) Number of observations

0:357 (0 :232)

43 131 :33 0:185

43 271:90 0 :263

43 275 :52 0:207

2 statistic for Wald Testc

Pseudo R-squared

The sample includes all 43 non-trademarked equity-linked and derivative securities b etween 1985 and 2001.
a

Estimates of the change in the probability that the security is imitated given an innitesimal change in the associated continuous variable, or a change of 0 to 1 in the associated dummy variable; estimated at the sample mean. The standard errors are shown below the estimate, in parenthesis, and are obtained using a Huber-White covariance estimator, robust to heteroskedasticity and correlation within-product groups. Estimates followed by

are signicant to the 0.01 level; by

to the 0.05 level, and by

to the 0.1 level.

Category is 1 if generation number is 1; 2 if generation number is between 2 and 6; and 3 otherwise. The null hypothesis is that all coecients are equal to zero.

83

5000 4500
MIPS

4000 3500 US$ Millions 3000 2500 2000 1500 1000 500 0
93-IV 94-II 94-IV 95-II 95-IV 96-II

QUIPS

TOPRS

QUICS

96-IV

97-II

97-IV

Quarters

98-II

98-IV

99-II

99-IV

00-II

00-IV

Figure 1: Principal Underwritten Using Tax-Deductible Preferred Products

84

3000

2500

PERCs

2000 US $ Millions

1500

1000 DECS 500

STRYPES

PEPS

0
91-II 91-IV 92-II 92-IV 93-II 93-IV 94-II 94-IV 95-II

Quarters

95-IV 96-II 96-IV

97-II

97-IV

98-II

98-IV

99-II

99-IV

00-II

00-IV

Figure 2: Principal Underwritten Using Convertible Preferred Products

85

10000

Morgan Stanley
8000

6000

$ Millions
4000 2000

Dean-Witter Merrill Lynch

0 0 500 1000 1500 2000 2500

Days

Figure 3: Cumulative Principal Underwritten, by banks, using PERCS (1st generation)

86

3000

2500

Salomon
2000

US $ Millions

1500

1000

Lehman
500

0 0 500 1000 1500 2000 2500

Days

Figure 4: Cumulative Principal Underwritten, by banks, using DECS (3rd generation)

87

1000 MSDW

DLJ 750 Lehman

US $ Millions

500

NA

250 Salomon LAZ BAIRD 0 0 200 400 600 800 1000 1200 1400 1600 1800

Days

Figure 5: Cumulative Principal Underwritten, by banks, using Trust-Originated Convertible Preferreds (14th generation)

88

2000

1600

Morgan Stanley

US $ Millions

1200

800

400

Goldman, Sachs

0 0 500 1000 1500 2000 2500 3000 3500

Days

Figure 6: Cumulative Principal Underwritten, by banks, using PERLS (1st generation)

89

120 Merrill Lynch 100

80

US$ Millions

Goldman, Sachs

60

40

Lehman

20

0 0 200 400 600 800 1000 1200 1400 1600

Days

Figure 7: Cumulative Principal Underwritten, by banks, using SMART Notes (4th generation)

90

Firm

1 Security

...

...

1 Banker Value of Variety ui1

...

#(G 1)

ui2 ...

ui#(G1)

uij

...

uiJ

Figure 8: The decision tree of an issuer of a corporate security. The issuer chooses a variety, which is given by a pair of a security type and a banker.

91

11 10

Advantage: s innovator / s imitator

9 8 7 6 5 4 3 2 1 0 2 4 6 8 10 12 14 16 18

Gen. 7

Generation 2

Generation 1

Years

Figure 9: Estimated Advantage to the Innovator over Time and Generations (nested logit demand; t = 8)

92

11 10

Advantage: sinnovator / simitator

9 8 7 6 5 4 3 2 1 0 2 4 6 8 10 12 14 16 18

Gen. 7

Generation 2

Generation 1

Years

Figure 10: Estimated Advantage to the Innovator over Time and Generations (nested logit; t = 11 )

93

13

Advantage: sinnovator / simitator

11

Gen. 7
0 2 4

Generation 2

Generation 1

1 6 8 10 12 14 16 18

Years

Figure 11: Estimated Advantage of the Innovator over Time and Generations (nested logit; t = 12)

94

Advantage: s innovator / simitator

Generation 1
4

Gen. 7

Generation 2

1 0 2 4 6 8 10 12

Years

Figure 12: Estimated Advantage to the Innovator over Time and Generations (nested logit; t = 16 )

95

11 10

Advantage: sinnovator / simitator

9 8 7 6 5 4 3 2 1 0 2 4 6 8 10 12 14 16 18

Generation 1 Generation 2

Gen. 7

Years

Figure 13: Estimated Advantage to the Innovator over Time and Generations (nested logit with issuer heterogeneity; t = 8 )

96